Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,749 +1,749 @@
|
|
1 |
-
import gradio as gr
|
2 |
-
import sqlite3
|
3 |
-
import json
|
4 |
-
import pandas as pd
|
5 |
-
from openai import OpenAI
|
6 |
-
import traceback
|
7 |
-
from typing import Dict, List, Tuple, Any
|
8 |
-
import re
|
9 |
-
from datetime import datetime
|
10 |
-
import threading
|
11 |
-
import queue
|
12 |
-
import html
|
13 |
-
import sys
|
14 |
-
import os
|
15 |
-
|
16 |
-
# Force stdout to use UTF-8 encoding to handle Unicode characters
|
17 |
-
if sys.stdout.encoding != 'utf-8':
|
18 |
-
sys.stdout = open(sys.stdout.fileno(), mode='w', encoding='utf-8', buffering=1)
|
19 |
-
|
20 |
-
class DatabaseQueryAgent:
|
21 |
-
def __init__(self, db_path: str = "innovativeskills.db"):
|
22 |
-
self.db_path = db_path
|
23 |
-
self.client = None
|
24 |
-
|
25 |
-
# Available models
|
26 |
-
self.models = {
|
27 |
-
"llama": "meta-llama/llama-3.
|
28 |
-
"mistral": "mistralai/mistral-7b-instruct:free",
|
29 |
-
"gemma": "google/gemma-2-9b-it:free" # Verification model
|
30 |
-
}
|
31 |
-
|
32 |
-
# Initialize database connection
|
33 |
-
self.init_db_connection()
|
34 |
-
|
35 |
-
def init_db_connection(self):
|
36 |
-
"""Initialize database connection with UTF-8 encoding"""
|
37 |
-
try:
|
38 |
-
conn = sqlite3.connect(self.db_path, check_same_thread=False)
|
39 |
-
conn.execute("PRAGMA encoding = 'UTF-8';")
|
40 |
-
cursor = conn.cursor()
|
41 |
-
|
42 |
-
# Load table metadata
|
43 |
-
self.table_metadata = self.get_table_metadata(conn, cursor)
|
44 |
-
self.column_metadata = self.get_column_metadata(conn, cursor)
|
45 |
-
self.actual_schema = self.get_actual_schema(conn, cursor)
|
46 |
-
|
47 |
-
conn.close()
|
48 |
-
|
49 |
-
except Exception as e:
|
50 |
-
print(f"Database initialization error: {e}")
|
51 |
-
self.table_metadata = {}
|
52 |
-
self.column_metadata = {}
|
53 |
-
self.actual_schema = {}
|
54 |
-
|
55 |
-
def get_db_connection(self):
|
56 |
-
"""Get a new database connection with UTF-8 encoding"""
|
57 |
-
conn = sqlite3.connect(self.db_path, check_same_thread=False)
|
58 |
-
conn.execute("PRAGMA encoding = 'UTF-8';")
|
59 |
-
return conn
|
60 |
-
|
61 |
-
def get_actual_schema(self, conn, cursor) -> Dict:
|
62 |
-
"""Get actual database schema"""
|
63 |
-
try:
|
64 |
-
cursor.execute("SELECT name FROM sqlite_master WHERE type='table' AND name NOT LIKE 'sqlite_%'")
|
65 |
-
tables = [row[0] for row in cursor.fetchall()]
|
66 |
-
schema = {}
|
67 |
-
for table in tables:
|
68 |
-
cursor.execute(f"PRAGMA table_info({table})")
|
69 |
-
columns = cursor.fetchall()
|
70 |
-
try:
|
71 |
-
cursor.execute(f"SELECT * FROM {table} LIMIT 3")
|
72 |
-
sample_data = cursor.fetchall()
|
73 |
-
except Exception:
|
74 |
-
sample_data = []
|
75 |
-
try:
|
76 |
-
cursor.execute(f"SELECT COUNT(*) FROM {table}")
|
77 |
-
row_count = cursor.fetchone()[0]
|
78 |
-
except Exception:
|
79 |
-
row_count = 0
|
80 |
-
schema[table] = {
|
81 |
-
'columns': [{'name': col[1], 'type': col[2], 'notnull': col[3], 'pk': col[5]} for col in columns],
|
82 |
-
'sample_data': sample_data,
|
83 |
-
'row_count': row_count
|
84 |
-
}
|
85 |
-
return schema
|
86 |
-
except Exception as e:
|
87 |
-
print(f"Error getting actual schema: {e}")
|
88 |
-
return {}
|
89 |
-
|
90 |
-
def get_table_metadata(self, conn, cursor) -> Dict:
|
91 |
-
"""Get table metadata"""
|
92 |
-
try:
|
93 |
-
query = """
|
94 |
-
SELECT table_name, domain, description, row_count
|
95 |
-
FROM table_catalog
|
96 |
-
WHERE table_name NOT IN ('table_catalog', 'column_catalog')
|
97 |
-
"""
|
98 |
-
results = cursor.execute(query).fetchall()
|
99 |
-
metadata = {}
|
100 |
-
for table_name, domain, description, row_count in results:
|
101 |
-
metadata[table_name] = {
|
102 |
-
'domain': domain,
|
103 |
-
'description': description,
|
104 |
-
'row_count': row_count
|
105 |
-
}
|
106 |
-
return metadata
|
107 |
-
except Exception as e:
|
108 |
-
print(f"Error loading table metadata: {e}")
|
109 |
-
return {}
|
110 |
-
|
111 |
-
def get_column_metadata(self, conn, cursor) -> Dict:
|
112 |
-
"""Get column metadata"""
|
113 |
-
try:
|
114 |
-
query = """
|
115 |
-
SELECT table_name, column_name, data_type, is_foreign_key, references_table, description
|
116 |
-
FROM column_catalog
|
117 |
-
"""
|
118 |
-
results = cursor.execute(query).fetchall()
|
119 |
-
metadata = {}
|
120 |
-
for table_name, column_name, data_type, is_fk, ref_table, description in results:
|
121 |
-
if table_name not in metadata:
|
122 |
-
metadata[table_name] = []
|
123 |
-
metadata[table_name].append({
|
124 |
-
'name': column_name,
|
125 |
-
'type': data_type,
|
126 |
-
'is_foreign_key': bool(is_fk),
|
127 |
-
'references': ref_table,
|
128 |
-
'description': description
|
129 |
-
})
|
130 |
-
return metadata
|
131 |
-
except Exception as e:
|
132 |
-
print(f"Error loading column metadata: {e}")
|
133 |
-
return {}
|
134 |
-
|
135 |
-
def setup_client(self, api_key: str):
|
136 |
-
"""Setup OpenRouter client"""
|
137 |
-
self.client = OpenAI(
|
138 |
-
base_url="https://openrouter.ai/api/v1",
|
139 |
-
api_key=api_key,
|
140 |
-
)
|
141 |
-
|
142 |
-
def get_relevant_tables_for_query(self, query: str) -> str:
|
143 |
-
"""Analyze query and return relevant table info"""
|
144 |
-
query_lower = query.lower()
|
145 |
-
relevant_tables = []
|
146 |
-
keywords = {
|
147 |
-
'customer': ['customer', 'client', 'buyer', 'user'],
|
148 |
-
'order': ['order', 'purchase', 'transaction', 'sale'],
|
149 |
-
'product': ['product', 'item', 'inventory', 'stock'],
|
150 |
-
'employee': ['employee', 'staff', 'worker', 'personnel'],
|
151 |
-
'patient': ['patient', 'medical', 'health'],
|
152 |
-
'student': ['student', 'enrollment', 'grade', 'course'],
|
153 |
-
'supplier': ['supplier', 'vendor', 'provider'],
|
154 |
-
'shipping': ['shipping', 'delivery', 'logistics'],
|
155 |
-
'payment': ['payment', 'invoice', 'billing'],
|
156 |
-
'account': ['account', 'financial', 'balance']
|
157 |
-
}
|
158 |
-
for concept, search_terms in keywords.items():
|
159 |
-
if any(term in query_lower for term in search_terms):
|
160 |
-
for table_name in self.actual_schema.keys():
|
161 |
-
table_lower = table_name.lower()
|
162 |
-
if any(term in table_lower for term in search_terms):
|
163 |
-
if table_name not in relevant_tables:
|
164 |
-
relevant_tables.append(table_name)
|
165 |
-
if not relevant_tables:
|
166 |
-
relevant_tables = [name for name, info in self.actual_schema.items()
|
167 |
-
if info['row_count'] > 10][:10]
|
168 |
-
schema_info = ""
|
169 |
-
for table in relevant_tables[:15]:
|
170 |
-
if table in self.actual_schema:
|
171 |
-
info = self.actual_schema[table]
|
172 |
-
columns_str = ", ".join([f"{col['name']}({col['type']})" for col in info['columns']])
|
173 |
-
schema_info += f"\nTable: {table}\n"
|
174 |
-
schema_info += f" Columns: {columns_str}\n"
|
175 |
-
schema_info += f" Rows: {info['row_count']}\n"
|
176 |
-
if table in self.table_metadata:
|
177 |
-
meta = self.table_metadata[table]
|
178 |
-
schema_info += f" Domain: {meta['domain']}\n"
|
179 |
-
schema_info += f" Description: {meta['description']}\n"
|
180 |
-
if info['sample_data']:
|
181 |
-
schema_info += f" Sample: {info['sample_data'][0] if info['sample_data'] else 'No data'}\n"
|
182 |
-
return schema_info
|
183 |
-
|
184 |
-
def get_system_prompt(self, user_query: str) -> str:
|
185 |
-
"""Generate system prompt with actual schema"""
|
186 |
-
relevant_schema = self.get_relevant_tables_for_query(user_query)
|
187 |
-
return f"""You are an intelligent database query agent that specializes in identifying relevant tables and generating accurate SQL queries.
|
188 |
-
|
189 |
-
DATABASE SCHEMA INFORMATION:
|
190 |
-
{relevant_schema}
|
191 |
-
|
192 |
-
CRITICAL SQL RULES:
|
193 |
-
1. NEVER use reserved words as table aliases (like 'to', 'from', 'where', 'select', etc.)
|
194 |
-
2. Use descriptive aliases like 'cust', 'ord', 'prod' instead
|
195 |
-
3. Only JOIN tables if you can identify a logical relationship between them
|
196 |
-
4. If no clear JOIN relationship exists, use separate SELECT statements or UNION
|
197 |
-
5. Always use the EXACT column names shown in the schema
|
198 |
-
6. Do not assume foreign key relationships unless explicitly shown
|
199 |
-
|
200 |
-
CRITICAL: You MUST respond with ONLY a valid JSON object. No markdown, no explanations outside the JSON.
|
201 |
-
|
202 |
-
Your response must be exactly in this JSON format:
|
203 |
-
{{
|
204 |
-
"analysis": "Brief analysis of the query and table selection reasoning",
|
205 |
-
"identified_tables": ["table1", "table2", "table3"],
|
206 |
-
"domains_involved": ["domain1", "domain2"],
|
207 |
-
"sql_query": "SELECT ... FROM ... WHERE ...",
|
208 |
-
"explanation": "Step-by-step explanation of the query logic",
|
209 |
-
"confidence": 0.95,
|
210 |
-
"alternative_queries": ["Alternative SQL if applicable"]
|
211 |
-
}}
|
212 |
-
|
213 |
-
IMPORTANT RULES:
|
214 |
-
1. Respond with ONLY valid JSON - no markdown formatting
|
215 |
-
2. Use ONLY the actual table names shown in the schema above
|
216 |
-
3. Use ONLY the actual column names shown in the schema above
|
217 |
-
4. Generate syntactically correct SQL queries with proper aliases
|
218 |
-
5. Focus on tables that actually exist and have relevant data
|
219 |
-
6. Include confidence scores between 0.0 and 1.0
|
220 |
-
7. Provide clear explanations
|
221 |
-
8. Ensure table names in 'identified_tables' match those used in 'sql_query'
|
222 |
-
9. Check that columns referenced in SQL actually exist in the tables
|
223 |
-
10. If no perfect match exists, choose the closest relevant tables and explain the compromise
|
224 |
-
11. Avoid reserved word aliases like 'to', 'from', 'order', 'select'
|
225 |
-
|
226 |
-
QUERY ANALYSIS GUIDELINES:
|
227 |
-
- For customer/order queries: Look for tables with customer-related or order-related names and columns
|
228 |
-
- For employee queries: Look for tables with employee, staff, or HR-related names
|
229 |
-
- For product queries: Look for tables with product, inventory, or item-related names
|
230 |
-
- Always verify column names exist before using them in SQL
|
231 |
-
- Use proper JOIN syntax when combining tables, but only if logical relationships exist
|
232 |
-
- Include appropriate WHERE clauses when filtering is implied
|
233 |
-
- If unsure about relationships, prefer simpler queries or multiple separate queries"""
|
234 |
-
|
235 |
-
def extract_json_from_response(self, response_text: str) -> Dict:
|
236 |
-
"""Extract JSON from response text"""
|
237 |
-
try:
|
238 |
-
return json.loads(response_text)
|
239 |
-
except json.JSONDecodeError:
|
240 |
-
json_pattern = r'```json\s*(.*?)\s*```'
|
241 |
-
json_match = re.search(json_pattern, response_text, re.DOTALL)
|
242 |
-
if json_match:
|
243 |
-
try:
|
244 |
-
return json.loads(json_match.group(1))
|
245 |
-
except json.JSONDecodeError:
|
246 |
-
pass
|
247 |
-
json_pattern = r'\{.*\}'
|
248 |
-
json_match = re.search(json_pattern, response_text, re.DOTALL)
|
249 |
-
if json_match:
|
250 |
-
try:
|
251 |
-
return json.loads(json_match.group(0))
|
252 |
-
except json.JSONDecodeError:
|
253 |
-
pass
|
254 |
-
return self.create_fallback_response(response_text)
|
255 |
-
|
256 |
-
def create_fallback_response(self, response_text: str) -> Dict:
|
257 |
-
"""Create a fallback response when JSON parsing fails"""
|
258 |
-
sql_pattern = r'SELECT.*?(?:;|$)'
|
259 |
-
sql_match = re.search(sql_pattern, response_text, re.IGNORECASE | re.DOTALL)
|
260 |
-
sql_query = sql_match.group(0).strip(';') if sql_match else ""
|
261 |
-
identified_tables = [table_name for table_name in self.actual_schema.keys()
|
262 |
-
if table_name.lower() in response_text.lower()]
|
263 |
-
domains_involved = [self.table_metadata[table]['domain'] for table in identified_tables
|
264 |
-
if table in self.table_metadata and self.table_metadata[table]['domain'] not in domains_involved]
|
265 |
-
return {
|
266 |
-
"analysis": "Fallback analysis from unparseable response",
|
267 |
-
"identified_tables": identified_tables[:5],
|
268 |
-
"domains_involved": domains_involved[:3],
|
269 |
-
"sql_query": sql_query,
|
270 |
-
"explanation": "Response could not be parsed as JSON, extracted information where possible",
|
271 |
-
"confidence": 0.5,
|
272 |
-
"alternative_queries": []
|
273 |
-
}
|
274 |
-
|
275 |
-
def validate_sql_query(self, sql_query: str, identified_tables: List[str]) -> Tuple[bool, str]:
|
276 |
-
"""Validate SQL query against schema"""
|
277 |
-
try:
|
278 |
-
if not sql_query.strip():
|
279 |
-
return False, "Empty SQL query"
|
280 |
-
for table in identified_tables:
|
281 |
-
if table not in self.actual_schema:
|
282 |
-
return False, f"Table '{table}' does not exist in database"
|
283 |
-
sql_upper = sql_query.upper()
|
284 |
-
if not sql_upper.strip().startswith('SELECT'):
|
285 |
-
return False, "Only SELECT queries are allowed"
|
286 |
-
reserved_words = ['TO', 'FROM', 'WHERE', 'SELECT', 'ORDER', 'GROUP', 'HAVING', 'UNION', 'JOIN', 'ON']
|
287 |
-
alias_pattern = r'(?:FROM|JOIN)\s+(\w+)\s+(\w+)'
|
288 |
-
aliases = re.findall(alias_pattern, sql_query, re.IGNORECASE)
|
289 |
-
for table, alias in aliases:
|
290 |
-
if alias.upper() in reserved_words:
|
291 |
-
return False, f"Cannot use reserved word '{alias}' as table alias"
|
292 |
-
for table in identified_tables:
|
293 |
-
if table in sql_query:
|
294 |
-
table_info = self.actual_schema[table]
|
295 |
-
available_columns = [col['name'] for col in table_info['columns']]
|
296 |
-
column_patterns = [
|
297 |
-
rf'{re.escape(table)}\.(\w+)',
|
298 |
-
rf'\b(\w+)\.(\w+)',
|
299 |
-
rf'SELECT\s+([^FROM]+)'
|
300 |
-
]
|
301 |
-
for pattern in column_patterns:
|
302 |
-
matches = re.findall(pattern, sql_query, re.IGNORECASE)
|
303 |
-
for match in matches:
|
304 |
-
if isinstance(match, tuple):
|
305 |
-
column = match[1] if len(match) == 2 else match[0] if match else ''
|
306 |
-
else:
|
307 |
-
column = match
|
308 |
-
if column.upper() in ['*', 'COUNT', 'SUM', 'AVG', 'MAX', 'MIN', 'DISTINCT']:
|
309 |
-
continue
|
310 |
-
if column and column not in available_columns and f'{table}.{column}' in sql_query:
|
311 |
-
return False, f"Column '{column}' does not exist in table '{table}'"
|
312 |
-
return True, "Query validation passed"
|
313 |
-
except Exception as e:
|
314 |
-
return False, f"Validation error: {str(e)}"
|
315 |
-
|
316 |
-
def call_model(self, model_key: str, prompt: str, user_query: str) -> Dict:
|
317 |
-
"""Call specific model with prompt"""
|
318 |
-
try:
|
319 |
-
messages = [
|
320 |
-
{"role": "system", "content": prompt},
|
321 |
-
{"role": "user", "content": f"Query: {user_query}\n\nRespond with ONLY a valid JSON object following the exact format specified in the system prompt."}
|
322 |
-
]
|
323 |
-
completion = self.client.chat.completions.create(
|
324 |
-
model=self.models[model_key],
|
325 |
-
messages=messages,
|
326 |
-
temperature=0.1,
|
327 |
-
max_tokens=2000
|
328 |
-
)
|
329 |
-
response = completion.choices[0].message.content.strip()
|
330 |
-
parsed_response = self.extract_json_from_response(response)
|
331 |
-
sql_query = parsed_response.get('sql_query', '')
|
332 |
-
identified_tables = parsed_response.get('identified_tables', [])
|
333 |
-
if sql_query:
|
334 |
-
is_valid, validation_message = self.validate_sql_query(sql_query, identified_tables)
|
335 |
-
parsed_response['sql_validation'] = {
|
336 |
-
'is_valid': is_valid,
|
337 |
-
'message': validation_message
|
338 |
-
}
|
339 |
-
return {
|
340 |
-
"success": True,
|
341 |
-
"response": parsed_response,
|
342 |
-
"raw_response": response,
|
343 |
-
"model": model_key
|
344 |
-
}
|
345 |
-
except Exception as e:
|
346 |
-
return {
|
347 |
-
"success": False,
|
348 |
-
"error": str(e),
|
349 |
-
"model": model_key
|
350 |
-
}
|
351 |
-
|
352 |
-
def verify_response(self, api_key: str, original_query: str, llama_response: Dict, mistral_response: Dict) -> Dict:
|
353 |
-
"""Use Gemma to verify responses"""
|
354 |
-
self.setup_client(api_key)
|
355 |
-
relevant_schema = self.get_relevant_tables_for_query(original_query)
|
356 |
-
verification_prompt = f"""You are a database query verification expert. You have access to the actual database schema and must verify responses against it.
|
357 |
-
|
358 |
-
ACTUAL DATABASE SCHEMA:
|
359 |
-
{relevant_schema}
|
360 |
-
|
361 |
-
ORIGINAL QUERY: {original_query}
|
362 |
-
|
363 |
-
LLAMA RESPONSE: {json.dumps(llama_response.get('response', {}), indent=2)}
|
364 |
-
|
365 |
-
MISTRAL RESPONSE: {json.dumps(mistral_response.get('response', {}), indent=2)}
|
366 |
-
|
367 |
-
Verify these responses against the ACTUAL schema above. Check:
|
368 |
-
1. Do the table names actually exist in the schema?
|
369 |
-
2. Do the column names actually exist in those tables?
|
370 |
-
3. Are the table selections appropriate for the query?
|
371 |
-
4. Is the SQL syntax correct?
|
372 |
-
5. Are table aliases proper (not reserved words)?
|
373 |
-
|
374 |
-
Respond with ONLY a valid JSON object:
|
375 |
-
{{
|
376 |
-
"verification_summary": "Overall assessment based on actual schema",
|
377 |
-
"table_selection_accuracy": "Assessment of table choices against actual schema",
|
378 |
-
"sql_correctness": "SQL syntax and schema validation",
|
379 |
-
"consistency_check": "Comparison between responses",
|
380 |
-
"recommended_response": "llama, mistral, or neither",
|
381 |
-
"confidence_score": 0.85,
|
382 |
-
"suggested_improvements": ["improvement1", "improvement2"],
|
383 |
-
"potential_issues": ["issue1", "issue2"],
|
384 |
-
"schema_compliance": "Assessment of how well responses match actual schema"
|
385 |
-
}}"""
|
386 |
-
return self.call_model("gemma", verification_prompt, "Verify the above responses against the actual database schema.")
|
387 |
-
|
388 |
-
def execute_query_in_thread(self, sql_query: str, result_queue: queue.Queue):
|
389 |
-
"""Execute SQL query in a thread"""
|
390 |
-
try:
|
391 |
-
if not sql_query.strip().upper().startswith('SELECT'):
|
392 |
-
result_queue.put((False, "Only SELECT queries are allowed"))
|
393 |
-
return
|
394 |
-
sql_query = sql_query.strip().rstrip(';')
|
395 |
-
conn = self.get_db_connection()
|
396 |
-
try:
|
397 |
-
df = pd.read_sql_query(sql_query, conn)
|
398 |
-
result_queue.put((True, df))
|
399 |
-
except Exception as e:
|
400 |
-
result_queue.put((False, str(e)))
|
401 |
-
finally:
|
402 |
-
conn.close()
|
403 |
-
except Exception as e:
|
404 |
-
result_queue.put((False, f"Query execution error: {str(e)}"))
|
405 |
-
|
406 |
-
def execute_query(self, sql_query: str) -> Tuple[bool, Any]:
|
407 |
-
"""Execute SQL query using thread-safe approach"""
|
408 |
-
try:
|
409 |
-
result_queue = queue.Queue()
|
410 |
-
thread = threading.Thread(
|
411 |
-
target=self.execute_query_in_thread,
|
412 |
-
args=(sql_query, result_queue)
|
413 |
-
)
|
414 |
-
thread.start()
|
415 |
-
thread.join(timeout=30)
|
416 |
-
if thread.is_alive():
|
417 |
-
return False, "Query execution timed out"
|
418 |
-
if not result_queue.empty():
|
419 |
-
return result_queue.get()
|
420 |
-
else:
|
421 |
-
return False, "No result returned from query execution"
|
422 |
-
except Exception as e:
|
423 |
-
return False, f"Execution error: {str(e)}"
|
424 |
-
|
425 |
-
def process_query(self, api_key: str, user_query: str) -> Dict:
|
426 |
-
"""Process user query"""
|
427 |
-
if not api_key:
|
428 |
-
return {"error": "Please provide OpenRouter API key"}
|
429 |
-
try:
|
430 |
-
self.setup_client(api_key)
|
431 |
-
system_prompt = self.get_system_prompt(user_query)
|
432 |
-
llama_result = self.call_model("llama", system_prompt, user_query)
|
433 |
-
mistral_result = self.call_model("mistral", system_prompt, user_query)
|
434 |
-
verification_result = self.verify_response(api_key, user_query, llama_result, mistral_result)
|
435 |
-
execution_results = {}
|
436 |
-
for model_name, result in [("llama", llama_result), ("mistral", mistral_result)]:
|
437 |
-
if result.get("success") and result.get("response", {}).get("sql_query"):
|
438 |
-
sql_query = result["response"]["sql_query"]
|
439 |
-
validation_info = result["response"].get("sql_validation", {})
|
440 |
-
if sql_query.strip():
|
441 |
-
if validation_info.get("is_valid", True):
|
442 |
-
success, data = self.execute_query(sql_query)
|
443 |
-
execution_results[model_name] = {
|
444 |
-
"success": success,
|
445 |
-
"data": data.to_dict('records') if success and isinstance(data, pd.DataFrame) else str(data),
|
446 |
-
"row_count": len(data) if success and isinstance(data, pd.DataFrame) else 0,
|
447 |
-
"sql_query": sql_query,
|
448 |
-
"validation": validation_info
|
449 |
-
}
|
450 |
-
else:
|
451 |
-
execution_results[model_name] = {
|
452 |
-
"success": False,
|
453 |
-
"data": f"Query validation failed: {validation_info.get('message', 'Unknown error')}",
|
454 |
-
"row_count": 0,
|
455 |
-
"sql_query": sql_query,
|
456 |
-
"validation": validation_info
|
457 |
-
}
|
458 |
-
else:
|
459 |
-
execution_results[model_name] = {
|
460 |
-
"success": False,
|
461 |
-
"data": "No SQL query generated",
|
462 |
-
"row_count": 0,
|
463 |
-
"sql_query": "",
|
464 |
-
"validation": {"is_valid": False, "message": "Empty query"}
|
465 |
-
}
|
466 |
-
else:
|
467 |
-
execution_results[model_name] = {
|
468 |
-
"success": False,
|
469 |
-
"data": "Model failed to generate response",
|
470 |
-
"row_count": 0,
|
471 |
-
"sql_query": "",
|
472 |
-
"validation": {"is_valid": False, "message": "Model error"}
|
473 |
-
}
|
474 |
-
return {
|
475 |
-
"llama_response": llama_result,
|
476 |
-
"mistral_response": mistral_result,
|
477 |
-
"verification": verification_result,
|
478 |
-
"execution_results": execution_results,
|
479 |
-
"timestamp": datetime.now().isoformat(),
|
480 |
-
"schema_info": self.get_relevant_tables_for_query(user_query)
|
481 |
-
}
|
482 |
-
except Exception as e:
|
483 |
-
return {"error": f"Processing error: {str(e)}", "traceback": traceback.format_exc()}
|
484 |
-
|
485 |
-
def response_to_markdown(response_dict: Dict) -> str:
|
486 |
-
"""Convert model response to Markdown"""
|
487 |
-
if not response_dict.get("success", False):
|
488 |
-
return f"**Error**: {response_dict.get('error', 'Unknown error')}"
|
489 |
-
response = response_dict.get("response", {})
|
490 |
-
markdown = "**Query Analysis Results**\n\n"
|
491 |
-
markdown += f"- **Analysis**: {response.get('analysis', 'N/A')}\n\n"
|
492 |
-
identified_tables = response.get('identified_tables', [])
|
493 |
-
markdown += f"- **Identified Tables**: {', '.join(identified_tables) if identified_tables else 'None'}\n\n"
|
494 |
-
domains_involved = response.get('domains_involved', [])
|
495 |
-
markdown += f"- **Domains Involved**: {', '.join(domains_involved) if domains_involved else 'None'}\n\n"
|
496 |
-
sql_query = response.get('sql_query', '')
|
497 |
-
if sql_query:
|
498 |
-
markdown += "- **SQL Query**:\n\n```sql\n" + sql_query + "\n```\n\n"
|
499 |
-
else:
|
500 |
-
markdown += "- **SQL Query**: None\n\n"
|
501 |
-
markdown += f"- **Explanation**: {response.get('explanation', 'N/A')}\n\n"
|
502 |
-
markdown += f"- **Confidence**: {response.get('confidence', 'N/A')}\n\n"
|
503 |
-
alternative_queries = response.get('alternative_queries', [])
|
504 |
-
if alternative_queries:
|
505 |
-
markdown += "- **Alternative Queries**:\n"
|
506 |
-
for query in alternative_queries:
|
507 |
-
markdown += f" - {query}\n"
|
508 |
-
else:
|
509 |
-
markdown += "- **Alternative Queries**: None\n"
|
510 |
-
validation = response.get('sql_validation', {})
|
511 |
-
if validation:
|
512 |
-
is_valid = validation.get('is_valid', False)
|
513 |
-
message = validation.get('message', 'N/A')
|
514 |
-
markdown += f"\n- **SQL Validation**: {'Passed' if is_valid else 'Failed'} - {message}\n"
|
515 |
-
return markdown
|
516 |
-
|
517 |
-
def verification_to_markdown(verification_dict: Dict) -> str:
|
518 |
-
"""Convert verification response to Markdown"""
|
519 |
-
if not verification_dict.get("success", False):
|
520 |
-
return f"**Error**: {verification_dict.get('error', 'Unknown error')}"
|
521 |
-
response = verification_dict.get("response", {})
|
522 |
-
markdown = "**Verification Results**\n\n"
|
523 |
-
markdown += f"- **Verification Summary**: {response.get('verification_summary', 'N/A')}\n\n"
|
524 |
-
markdown += f"- **Table Selection Accuracy**: {response.get('table_selection_accuracy', 'N/A')}\n\n"
|
525 |
-
markdown += f"- **SQL Correctness**: {response.get('sql_correctness', 'N/A')}\n\n"
|
526 |
-
markdown += f"- **Consistency Check**: {response.get('consistency_check', 'N/A')}\n\n"
|
527 |
-
markdown += f"- **Recommended Response**: {response.get('recommended_response', 'N/A')}\n\n"
|
528 |
-
markdown += f"- **Confidence Score**: {response.get('confidence_score', 'N/A')}\n\n"
|
529 |
-
suggested_improvements = response.get('suggested_improvements', [])
|
530 |
-
if suggested_improvements:
|
531 |
-
markdown += "- **Suggested Improvements**:\n"
|
532 |
-
for improvement in suggested_improvements:
|
533 |
-
markdown += f" - {improvement}\n"
|
534 |
-
else:
|
535 |
-
markdown += "- **Suggested Improvements**: None\n"
|
536 |
-
potential_issues = response.get('potential_issues', [])
|
537 |
-
if potential_issues:
|
538 |
-
markdown += "- **Potential Issues**:\n"
|
539 |
-
for issue in potential_issues:
|
540 |
-
markdown += f" - {issue}\n"
|
541 |
-
else:
|
542 |
-
markdown += "- **Potential Issues**: None\n"
|
543 |
-
markdown += f"- **Schema Compliance**: {response.get('schema_compliance', 'N/A')}\n"
|
544 |
-
return markdown
|
545 |
-
|
546 |
-
def create_gradio_interface():
|
547 |
-
"""Create Gradio interface"""
|
548 |
-
agent = DatabaseQueryAgent()
|
549 |
-
sample_queries = [
|
550 |
-
"Find all customers from customer tables",
|
551 |
-
"Show me employee information from HR tables",
|
552 |
-
"Get patient data from healthcare tables",
|
553 |
-
"List all products with their details",
|
554 |
-
"Find students enrolled in courses",
|
555 |
-
"Show financial transaction records",
|
556 |
-
"Get shipping information for deliveries",
|
557 |
-
"Find all suppliers and their information",
|
558 |
-
"Show retail store data",
|
559 |
-
"Get manufacturing production records"
|
560 |
-
]
|
561 |
-
|
562 |
-
def process_user_query(api_key, query):
|
563 |
-
"""Process query and return formatted results"""
|
564 |
-
if not query.strip():
|
565 |
-
return "Please enter a query", "", "", "", "", ""
|
566 |
-
results = agent.process_query(api_key, query)
|
567 |
-
if "error" in results:
|
568 |
-
return f"**Error**: {results['error']}", "", "", "", "", ""
|
569 |
-
|
570 |
-
# Format responses as Markdown
|
571 |
-
llama_markdown = response_to_markdown(results.get("llama_response", {}))
|
572 |
-
mistral_markdown = response_to_markdown(results.get("mistral_response", {}))
|
573 |
-
verification_markdown = verification_to_markdown(results.get("verification", {}))
|
574 |
-
|
575 |
-
# Format execution results
|
576 |
-
exec_results = results.get("execution_results", {})
|
577 |
-
execution_formatted = ""
|
578 |
-
for model, result in exec_results.items():
|
579 |
-
execution_formatted += f"\n=== {model.upper()} EXECUTION ===\n"
|
580 |
-
execution_formatted += f"SQL Query: {result.get('sql_query', 'N/A')}\n"
|
581 |
-
validation = result.get('validation', {})
|
582 |
-
if validation.get('is_valid'):
|
583 |
-
execution_formatted += f"β
Query Validation: PASSED\n"
|
584 |
-
else:
|
585 |
-
execution_formatted += f"β Query Validation: FAILED - {validation.get('message', 'Unknown error')}\n"
|
586 |
-
if result["success"]:
|
587 |
-
execution_formatted += f"β
Execution: Success! Retrieved {result['row_count']} rows\n"
|
588 |
-
if result["row_count"] > 0:
|
589 |
-
sample_data = result['data'][:3] if isinstance(result['data'], list) else []
|
590 |
-
execution_formatted += f"Sample data:\n{json.dumps(sample_data, indent=2)}\n"
|
591 |
-
else:
|
592 |
-
execution_formatted += "No data returned (empty result set)\n"
|
593 |
-
else:
|
594 |
-
execution_formatted += f"β Execution Error: {result['data']}\n"
|
595 |
-
execution_formatted += "\n"
|
596 |
-
if not execution_formatted:
|
597 |
-
execution_formatted = "No queries were executed. Check if valid SQL was generated."
|
598 |
-
|
599 |
-
schema_info = results.get('schema_info', 'No schema information available')
|
600 |
-
|
601 |
-
# Format summary as Markdown
|
602 |
-
verification_resp = results.get('verification', {}).get('response', {})
|
603 |
-
summary = f"""
|
604 |
-
**π QUERY ANALYSIS COMPLETE**
|
605 |
-
|
606 |
-
ββββββββββββββββββββββββ
|
607 |
-
|
608 |
-
**π Models Used**: Llama 3.1 8B, Mistral 7B, Gemma 2 9B (verification)
|
609 |
-
|
610 |
-
**β° Processed**: {results.get('timestamp', 'N/A')}
|
611 |
-
|
612 |
-
**π― Verification Summary**:
|
613 |
-
|
614 |
-
{verification_resp.get('verification_summary', 'N/A')}
|
615 |
-
|
616 |
-
**π‘ Recommended Model**: {verification_resp.get('recommended_response', 'N/A')}
|
617 |
-
|
618 |
-
**π Confidence**: {verification_resp.get('confidence_score', 'N/A')}
|
619 |
-
|
620 |
-
**ποΈ Schema Compliance**: {verification_resp.get('schema_compliance', 'N/A')}
|
621 |
-
|
622 |
-
**ποΈ Query Execution Status**:
|
623 |
-
|
624 |
-
{len(exec_results)} queries attempted
|
625 |
-
"""
|
626 |
-
|
627 |
-
return summary, llama_markdown, mistral_markdown, verification_markdown, execution_formatted, schema_info
|
628 |
-
|
629 |
-
with gr.Blocks(
|
630 |
-
title="Fixed Intelligent Database Query Agent",
|
631 |
-
theme=gr.themes.Soft(),
|
632 |
-
css="""
|
633 |
-
.gradio-container {
|
634 |
-
max-width: 1200px !important;
|
635 |
-
margin: 0 auto !important;
|
636 |
-
}
|
637 |
-
.result-box {
|
638 |
-
background-color: #f8f9fa;
|
639 |
-
border: 1px solid #dee2e6;
|
640 |
-
border-radius: 8px;
|
641 |
-
padding: 15px;
|
642 |
-
}
|
643 |
-
"""
|
644 |
-
) as interface:
|
645 |
-
gr.HTML("""
|
646 |
-
<div style="text-align: center; padding: 20px; background: linear-gradient(135deg, #667eea 0%, #764ba2 100%); color: white; border-radius: 10px; margin-bottom: 20px;">
|
647 |
-
<h1>π€ Fixed Intelligent Database Query Agent</h1>
|
648 |
-
<p>AI-powered agent that intelligently selects relevant tables from 100+ tables and generates optimized SQL queries</p>
|
649 |
-
<p><strong>Database:</strong> 100 tables across 10 business domains | <strong>Models:</strong> Llama 3.1 8B + Mistral 7B + Gemma 2 9B</p>
|
650 |
-
<p><strong>β
FIXED:</strong> Reserved Word Aliases | Enhanced Column Validation | Better SQL Syntax Checking</p>
|
651 |
-
</div>
|
652 |
-
""")
|
653 |
-
|
654 |
-
with gr.Row():
|
655 |
-
with gr.Column(scale=1):
|
656 |
-
api_key_input = gr.Textbox(
|
657 |
-
label="π OpenRouter API Key",
|
658 |
-
type="password",
|
659 |
-
placeholder="Enter your OpenRouter API key...",
|
660 |
-
info="Get your free API key from openrouter.ai"
|
661 |
-
)
|
662 |
-
query_input = gr.Textbox(
|
663 |
-
label="π¬ Database Query",
|
664 |
-
placeholder="Enter your natural language query...",
|
665 |
-
lines=3,
|
666 |
-
info="Example: 'Find all customers who placed orders in the last month'"
|
667 |
-
)
|
668 |
-
with gr.Row():
|
669 |
-
submit_btn = gr.Button("π Process Query", variant="primary", size="lg")
|
670 |
-
clear_btn = gr.Button("ποΈ Clear", variant="secondary")
|
671 |
-
gr.HTML("<h3>π Sample Test Queries</h3>")
|
672 |
-
sample_dropdown = gr.Dropdown(
|
673 |
-
choices=sample_queries,
|
674 |
-
label="Quick Test Examples",
|
675 |
-
info="Select a sample query to test the agent"
|
676 |
-
)
|
677 |
-
|
678 |
-
with gr.Column(scale=2):
|
679 |
-
summary_output = gr.Markdown(label="π Analysis Summary")
|
680 |
-
with gr.Tabs():
|
681 |
-
with gr.Tab("π¦ Llama 3.1 8B Response"):
|
682 |
-
llama_output = gr.Markdown(label="Llama Response")
|
683 |
-
with gr.Tab("π Mistral 7B Response"):
|
684 |
-
mistral_output = gr.Markdown(label="Mistral Response")
|
685 |
-
with gr.Tab("β
Verification (Gemma 2 9B)"):
|
686 |
-
verification_output = gr.Markdown(label="Verification Analysis")
|
687 |
-
with gr.Tab("ποΈ Query Execution Results"):
|
688 |
-
execution_output = gr.Textbox(
|
689 |
-
label="Database Execution Results",
|
690 |
-
lines=15,
|
691 |
-
max_lines=20,
|
692 |
-
elem_classes=["result-box"]
|
693 |
-
)
|
694 |
-
with gr.Tab("π Database Schema"):
|
695 |
-
schema_output = gr.Textbox(
|
696 |
-
label="Relevant Database Schema",
|
697 |
-
lines=15,
|
698 |
-
max_lines=20,
|
699 |
-
elem_classes=["result-box"]
|
700 |
-
)
|
701 |
-
|
702 |
-
submit_btn.click(
|
703 |
-
fn=process_user_query,
|
704 |
-
inputs=[api_key_input, query_input],
|
705 |
-
outputs=[summary_output, llama_output, mistral_output, verification_output, execution_output, schema_output]
|
706 |
-
)
|
707 |
-
clear_btn.click(
|
708 |
-
fn=lambda: ("", "", "", "", "", "", ""),
|
709 |
-
outputs=[query_input, summary_output, llama_output, mistral_output, verification_output, execution_output, schema_output]
|
710 |
-
)
|
711 |
-
sample_dropdown.change(
|
712 |
-
fn=lambda x: x,
|
713 |
-
inputs=[sample_dropdown],
|
714 |
-
outputs=[query_input]
|
715 |
-
)
|
716 |
-
gr.HTML("""
|
717 |
-
<div style="margin-top: 20px; padding: 15px; background-color: #f8f9fa; border-radius: 8px;">
|
718 |
-
<h3>π― How to Use</h3>
|
719 |
-
<ol>
|
720 |
-
<li><strong>API Key:</strong> Get a free API key from <a href="https://openrouter.ai" target="_blank">openrouter.ai</a></li>
|
721 |
-
<li><strong>Query:</strong> Enter your natural language database query</li>
|
722 |
-
<li><strong>Process:</strong> The agent will analyze your query across 100+ tables and generate optimized SQL</li>
|
723 |
-
<li><strong>Results:</strong> View responses from multiple AI models, verification analysis, and actual query execution results</li>
|
724 |
-
</ol>
|
725 |
-
<p><strong>Features:</strong></p>
|
726 |
-
<ul>
|
727 |
-
<li>π§ Multi-model AI analysis (Llama, Mistral, Gemma)</li>
|
728 |
-
<li>π Intelligent table selection from 100+ tables</li>
|
729 |
-
<li>β
SQL validation and syntax checking</li>
|
730 |
-
<li>ποΈ Real database query execution with results</li>
|
731 |
-
<li>π Cross-model verification and comparison</li>
|
732 |
-
</ul>
|
733 |
-
</div>
|
734 |
-
""")
|
735 |
-
|
736 |
-
return interface
|
737 |
-
|
738 |
-
def main():
|
739 |
-
"""Main function to launch the application"""
|
740 |
-
print("π Starting Intelligent Database Query Agent...")
|
741 |
-
print("π Loading database schema and metadata...")
|
742 |
-
interface = create_gradio_interface()
|
743 |
-
print("β
Database Query Agent Ready!")
|
744 |
-
print("π Access the interface at: http://localhost:7860")
|
745 |
-
print("π Don't forget to add your OpenRouter API key!")
|
746 |
-
interface.launch(share=True)
|
747 |
-
|
748 |
-
if __name__ == "__main__":
|
749 |
main()
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import sqlite3
|
3 |
+
import json
|
4 |
+
import pandas as pd
|
5 |
+
from openai import OpenAI
|
6 |
+
import traceback
|
7 |
+
from typing import Dict, List, Tuple, Any
|
8 |
+
import re
|
9 |
+
from datetime import datetime
|
10 |
+
import threading
|
11 |
+
import queue
|
12 |
+
import html
|
13 |
+
import sys
|
14 |
+
import os
|
15 |
+
|
16 |
+
# Force stdout to use UTF-8 encoding to handle Unicode characters
|
17 |
+
if sys.stdout.encoding != 'utf-8':
|
18 |
+
sys.stdout = open(sys.stdout.fileno(), mode='w', encoding='utf-8', buffering=1)
|
19 |
+
|
20 |
+
class DatabaseQueryAgent:
|
21 |
+
def __init__(self, db_path: str = "innovativeskills.db"):
|
22 |
+
self.db_path = db_path
|
23 |
+
self.client = None
|
24 |
+
|
25 |
+
# Available models
|
26 |
+
self.models = {
|
27 |
+
"llama": "meta-llama/llama-3.3-70b-instruct:free",
|
28 |
+
"mistral": "mistralai/mistral-7b-instruct:free",
|
29 |
+
"gemma": "google/gemma-2-9b-it:free" # Verification model
|
30 |
+
}
|
31 |
+
|
32 |
+
# Initialize database connection
|
33 |
+
self.init_db_connection()
|
34 |
+
|
35 |
+
def init_db_connection(self):
|
36 |
+
"""Initialize database connection with UTF-8 encoding"""
|
37 |
+
try:
|
38 |
+
conn = sqlite3.connect(self.db_path, check_same_thread=False)
|
39 |
+
conn.execute("PRAGMA encoding = 'UTF-8';")
|
40 |
+
cursor = conn.cursor()
|
41 |
+
|
42 |
+
# Load table metadata
|
43 |
+
self.table_metadata = self.get_table_metadata(conn, cursor)
|
44 |
+
self.column_metadata = self.get_column_metadata(conn, cursor)
|
45 |
+
self.actual_schema = self.get_actual_schema(conn, cursor)
|
46 |
+
|
47 |
+
conn.close()
|
48 |
+
|
49 |
+
except Exception as e:
|
50 |
+
print(f"Database initialization error: {e}")
|
51 |
+
self.table_metadata = {}
|
52 |
+
self.column_metadata = {}
|
53 |
+
self.actual_schema = {}
|
54 |
+
|
55 |
+
def get_db_connection(self):
|
56 |
+
"""Get a new database connection with UTF-8 encoding"""
|
57 |
+
conn = sqlite3.connect(self.db_path, check_same_thread=False)
|
58 |
+
conn.execute("PRAGMA encoding = 'UTF-8';")
|
59 |
+
return conn
|
60 |
+
|
61 |
+
def get_actual_schema(self, conn, cursor) -> Dict:
|
62 |
+
"""Get actual database schema"""
|
63 |
+
try:
|
64 |
+
cursor.execute("SELECT name FROM sqlite_master WHERE type='table' AND name NOT LIKE 'sqlite_%'")
|
65 |
+
tables = [row[0] for row in cursor.fetchall()]
|
66 |
+
schema = {}
|
67 |
+
for table in tables:
|
68 |
+
cursor.execute(f"PRAGMA table_info({table})")
|
69 |
+
columns = cursor.fetchall()
|
70 |
+
try:
|
71 |
+
cursor.execute(f"SELECT * FROM {table} LIMIT 3")
|
72 |
+
sample_data = cursor.fetchall()
|
73 |
+
except Exception:
|
74 |
+
sample_data = []
|
75 |
+
try:
|
76 |
+
cursor.execute(f"SELECT COUNT(*) FROM {table}")
|
77 |
+
row_count = cursor.fetchone()[0]
|
78 |
+
except Exception:
|
79 |
+
row_count = 0
|
80 |
+
schema[table] = {
|
81 |
+
'columns': [{'name': col[1], 'type': col[2], 'notnull': col[3], 'pk': col[5]} for col in columns],
|
82 |
+
'sample_data': sample_data,
|
83 |
+
'row_count': row_count
|
84 |
+
}
|
85 |
+
return schema
|
86 |
+
except Exception as e:
|
87 |
+
print(f"Error getting actual schema: {e}")
|
88 |
+
return {}
|
89 |
+
|
90 |
+
def get_table_metadata(self, conn, cursor) -> Dict:
|
91 |
+
"""Get table metadata"""
|
92 |
+
try:
|
93 |
+
query = """
|
94 |
+
SELECT table_name, domain, description, row_count
|
95 |
+
FROM table_catalog
|
96 |
+
WHERE table_name NOT IN ('table_catalog', 'column_catalog')
|
97 |
+
"""
|
98 |
+
results = cursor.execute(query).fetchall()
|
99 |
+
metadata = {}
|
100 |
+
for table_name, domain, description, row_count in results:
|
101 |
+
metadata[table_name] = {
|
102 |
+
'domain': domain,
|
103 |
+
'description': description,
|
104 |
+
'row_count': row_count
|
105 |
+
}
|
106 |
+
return metadata
|
107 |
+
except Exception as e:
|
108 |
+
print(f"Error loading table metadata: {e}")
|
109 |
+
return {}
|
110 |
+
|
111 |
+
def get_column_metadata(self, conn, cursor) -> Dict:
|
112 |
+
"""Get column metadata"""
|
113 |
+
try:
|
114 |
+
query = """
|
115 |
+
SELECT table_name, column_name, data_type, is_foreign_key, references_table, description
|
116 |
+
FROM column_catalog
|
117 |
+
"""
|
118 |
+
results = cursor.execute(query).fetchall()
|
119 |
+
metadata = {}
|
120 |
+
for table_name, column_name, data_type, is_fk, ref_table, description in results:
|
121 |
+
if table_name not in metadata:
|
122 |
+
metadata[table_name] = []
|
123 |
+
metadata[table_name].append({
|
124 |
+
'name': column_name,
|
125 |
+
'type': data_type,
|
126 |
+
'is_foreign_key': bool(is_fk),
|
127 |
+
'references': ref_table,
|
128 |
+
'description': description
|
129 |
+
})
|
130 |
+
return metadata
|
131 |
+
except Exception as e:
|
132 |
+
print(f"Error loading column metadata: {e}")
|
133 |
+
return {}
|
134 |
+
|
135 |
+
def setup_client(self, api_key: str):
|
136 |
+
"""Setup OpenRouter client"""
|
137 |
+
self.client = OpenAI(
|
138 |
+
base_url="https://openrouter.ai/api/v1",
|
139 |
+
api_key=api_key,
|
140 |
+
)
|
141 |
+
|
142 |
+
def get_relevant_tables_for_query(self, query: str) -> str:
|
143 |
+
"""Analyze query and return relevant table info"""
|
144 |
+
query_lower = query.lower()
|
145 |
+
relevant_tables = []
|
146 |
+
keywords = {
|
147 |
+
'customer': ['customer', 'client', 'buyer', 'user'],
|
148 |
+
'order': ['order', 'purchase', 'transaction', 'sale'],
|
149 |
+
'product': ['product', 'item', 'inventory', 'stock'],
|
150 |
+
'employee': ['employee', 'staff', 'worker', 'personnel'],
|
151 |
+
'patient': ['patient', 'medical', 'health'],
|
152 |
+
'student': ['student', 'enrollment', 'grade', 'course'],
|
153 |
+
'supplier': ['supplier', 'vendor', 'provider'],
|
154 |
+
'shipping': ['shipping', 'delivery', 'logistics'],
|
155 |
+
'payment': ['payment', 'invoice', 'billing'],
|
156 |
+
'account': ['account', 'financial', 'balance']
|
157 |
+
}
|
158 |
+
for concept, search_terms in keywords.items():
|
159 |
+
if any(term in query_lower for term in search_terms):
|
160 |
+
for table_name in self.actual_schema.keys():
|
161 |
+
table_lower = table_name.lower()
|
162 |
+
if any(term in table_lower for term in search_terms):
|
163 |
+
if table_name not in relevant_tables:
|
164 |
+
relevant_tables.append(table_name)
|
165 |
+
if not relevant_tables:
|
166 |
+
relevant_tables = [name for name, info in self.actual_schema.items()
|
167 |
+
if info['row_count'] > 10][:10]
|
168 |
+
schema_info = ""
|
169 |
+
for table in relevant_tables[:15]:
|
170 |
+
if table in self.actual_schema:
|
171 |
+
info = self.actual_schema[table]
|
172 |
+
columns_str = ", ".join([f"{col['name']}({col['type']})" for col in info['columns']])
|
173 |
+
schema_info += f"\nTable: {table}\n"
|
174 |
+
schema_info += f" Columns: {columns_str}\n"
|
175 |
+
schema_info += f" Rows: {info['row_count']}\n"
|
176 |
+
if table in self.table_metadata:
|
177 |
+
meta = self.table_metadata[table]
|
178 |
+
schema_info += f" Domain: {meta['domain']}\n"
|
179 |
+
schema_info += f" Description: {meta['description']}\n"
|
180 |
+
if info['sample_data']:
|
181 |
+
schema_info += f" Sample: {info['sample_data'][0] if info['sample_data'] else 'No data'}\n"
|
182 |
+
return schema_info
|
183 |
+
|
184 |
+
def get_system_prompt(self, user_query: str) -> str:
|
185 |
+
"""Generate system prompt with actual schema"""
|
186 |
+
relevant_schema = self.get_relevant_tables_for_query(user_query)
|
187 |
+
return f"""You are an intelligent database query agent that specializes in identifying relevant tables and generating accurate SQL queries.
|
188 |
+
|
189 |
+
DATABASE SCHEMA INFORMATION:
|
190 |
+
{relevant_schema}
|
191 |
+
|
192 |
+
CRITICAL SQL RULES:
|
193 |
+
1. NEVER use reserved words as table aliases (like 'to', 'from', 'where', 'select', etc.)
|
194 |
+
2. Use descriptive aliases like 'cust', 'ord', 'prod' instead
|
195 |
+
3. Only JOIN tables if you can identify a logical relationship between them
|
196 |
+
4. If no clear JOIN relationship exists, use separate SELECT statements or UNION
|
197 |
+
5. Always use the EXACT column names shown in the schema
|
198 |
+
6. Do not assume foreign key relationships unless explicitly shown
|
199 |
+
|
200 |
+
CRITICAL: You MUST respond with ONLY a valid JSON object. No markdown, no explanations outside the JSON.
|
201 |
+
|
202 |
+
Your response must be exactly in this JSON format:
|
203 |
+
{{
|
204 |
+
"analysis": "Brief analysis of the query and table selection reasoning",
|
205 |
+
"identified_tables": ["table1", "table2", "table3"],
|
206 |
+
"domains_involved": ["domain1", "domain2"],
|
207 |
+
"sql_query": "SELECT ... FROM ... WHERE ...",
|
208 |
+
"explanation": "Step-by-step explanation of the query logic",
|
209 |
+
"confidence": 0.95,
|
210 |
+
"alternative_queries": ["Alternative SQL if applicable"]
|
211 |
+
}}
|
212 |
+
|
213 |
+
IMPORTANT RULES:
|
214 |
+
1. Respond with ONLY valid JSON - no markdown formatting
|
215 |
+
2. Use ONLY the actual table names shown in the schema above
|
216 |
+
3. Use ONLY the actual column names shown in the schema above
|
217 |
+
4. Generate syntactically correct SQL queries with proper aliases
|
218 |
+
5. Focus on tables that actually exist and have relevant data
|
219 |
+
6. Include confidence scores between 0.0 and 1.0
|
220 |
+
7. Provide clear explanations
|
221 |
+
8. Ensure table names in 'identified_tables' match those used in 'sql_query'
|
222 |
+
9. Check that columns referenced in SQL actually exist in the tables
|
223 |
+
10. If no perfect match exists, choose the closest relevant tables and explain the compromise
|
224 |
+
11. Avoid reserved word aliases like 'to', 'from', 'order', 'select'
|
225 |
+
|
226 |
+
QUERY ANALYSIS GUIDELINES:
|
227 |
+
- For customer/order queries: Look for tables with customer-related or order-related names and columns
|
228 |
+
- For employee queries: Look for tables with employee, staff, or HR-related names
|
229 |
+
- For product queries: Look for tables with product, inventory, or item-related names
|
230 |
+
- Always verify column names exist before using them in SQL
|
231 |
+
- Use proper JOIN syntax when combining tables, but only if logical relationships exist
|
232 |
+
- Include appropriate WHERE clauses when filtering is implied
|
233 |
+
- If unsure about relationships, prefer simpler queries or multiple separate queries"""
|
234 |
+
|
235 |
+
def extract_json_from_response(self, response_text: str) -> Dict:
|
236 |
+
"""Extract JSON from response text"""
|
237 |
+
try:
|
238 |
+
return json.loads(response_text)
|
239 |
+
except json.JSONDecodeError:
|
240 |
+
json_pattern = r'```json\s*(.*?)\s*```'
|
241 |
+
json_match = re.search(json_pattern, response_text, re.DOTALL)
|
242 |
+
if json_match:
|
243 |
+
try:
|
244 |
+
return json.loads(json_match.group(1))
|
245 |
+
except json.JSONDecodeError:
|
246 |
+
pass
|
247 |
+
json_pattern = r'\{.*\}'
|
248 |
+
json_match = re.search(json_pattern, response_text, re.DOTALL)
|
249 |
+
if json_match:
|
250 |
+
try:
|
251 |
+
return json.loads(json_match.group(0))
|
252 |
+
except json.JSONDecodeError:
|
253 |
+
pass
|
254 |
+
return self.create_fallback_response(response_text)
|
255 |
+
|
256 |
+
def create_fallback_response(self, response_text: str) -> Dict:
|
257 |
+
"""Create a fallback response when JSON parsing fails"""
|
258 |
+
sql_pattern = r'SELECT.*?(?:;|$)'
|
259 |
+
sql_match = re.search(sql_pattern, response_text, re.IGNORECASE | re.DOTALL)
|
260 |
+
sql_query = sql_match.group(0).strip(';') if sql_match else ""
|
261 |
+
identified_tables = [table_name for table_name in self.actual_schema.keys()
|
262 |
+
if table_name.lower() in response_text.lower()]
|
263 |
+
domains_involved = [self.table_metadata[table]['domain'] for table in identified_tables
|
264 |
+
if table in self.table_metadata and self.table_metadata[table]['domain'] not in domains_involved]
|
265 |
+
return {
|
266 |
+
"analysis": "Fallback analysis from unparseable response",
|
267 |
+
"identified_tables": identified_tables[:5],
|
268 |
+
"domains_involved": domains_involved[:3],
|
269 |
+
"sql_query": sql_query,
|
270 |
+
"explanation": "Response could not be parsed as JSON, extracted information where possible",
|
271 |
+
"confidence": 0.5,
|
272 |
+
"alternative_queries": []
|
273 |
+
}
|
274 |
+
|
275 |
+
def validate_sql_query(self, sql_query: str, identified_tables: List[str]) -> Tuple[bool, str]:
|
276 |
+
"""Validate SQL query against schema"""
|
277 |
+
try:
|
278 |
+
if not sql_query.strip():
|
279 |
+
return False, "Empty SQL query"
|
280 |
+
for table in identified_tables:
|
281 |
+
if table not in self.actual_schema:
|
282 |
+
return False, f"Table '{table}' does not exist in database"
|
283 |
+
sql_upper = sql_query.upper()
|
284 |
+
if not sql_upper.strip().startswith('SELECT'):
|
285 |
+
return False, "Only SELECT queries are allowed"
|
286 |
+
reserved_words = ['TO', 'FROM', 'WHERE', 'SELECT', 'ORDER', 'GROUP', 'HAVING', 'UNION', 'JOIN', 'ON']
|
287 |
+
alias_pattern = r'(?:FROM|JOIN)\s+(\w+)\s+(\w+)'
|
288 |
+
aliases = re.findall(alias_pattern, sql_query, re.IGNORECASE)
|
289 |
+
for table, alias in aliases:
|
290 |
+
if alias.upper() in reserved_words:
|
291 |
+
return False, f"Cannot use reserved word '{alias}' as table alias"
|
292 |
+
for table in identified_tables:
|
293 |
+
if table in sql_query:
|
294 |
+
table_info = self.actual_schema[table]
|
295 |
+
available_columns = [col['name'] for col in table_info['columns']]
|
296 |
+
column_patterns = [
|
297 |
+
rf'{re.escape(table)}\.(\w+)',
|
298 |
+
rf'\b(\w+)\.(\w+)',
|
299 |
+
rf'SELECT\s+([^FROM]+)'
|
300 |
+
]
|
301 |
+
for pattern in column_patterns:
|
302 |
+
matches = re.findall(pattern, sql_query, re.IGNORECASE)
|
303 |
+
for match in matches:
|
304 |
+
if isinstance(match, tuple):
|
305 |
+
column = match[1] if len(match) == 2 else match[0] if match else ''
|
306 |
+
else:
|
307 |
+
column = match
|
308 |
+
if column.upper() in ['*', 'COUNT', 'SUM', 'AVG', 'MAX', 'MIN', 'DISTINCT']:
|
309 |
+
continue
|
310 |
+
if column and column not in available_columns and f'{table}.{column}' in sql_query:
|
311 |
+
return False, f"Column '{column}' does not exist in table '{table}'"
|
312 |
+
return True, "Query validation passed"
|
313 |
+
except Exception as e:
|
314 |
+
return False, f"Validation error: {str(e)}"
|
315 |
+
|
316 |
+
def call_model(self, model_key: str, prompt: str, user_query: str) -> Dict:
|
317 |
+
"""Call specific model with prompt"""
|
318 |
+
try:
|
319 |
+
messages = [
|
320 |
+
{"role": "system", "content": prompt},
|
321 |
+
{"role": "user", "content": f"Query: {user_query}\n\nRespond with ONLY a valid JSON object following the exact format specified in the system prompt."}
|
322 |
+
]
|
323 |
+
completion = self.client.chat.completions.create(
|
324 |
+
model=self.models[model_key],
|
325 |
+
messages=messages,
|
326 |
+
temperature=0.1,
|
327 |
+
max_tokens=2000
|
328 |
+
)
|
329 |
+
response = completion.choices[0].message.content.strip()
|
330 |
+
parsed_response = self.extract_json_from_response(response)
|
331 |
+
sql_query = parsed_response.get('sql_query', '')
|
332 |
+
identified_tables = parsed_response.get('identified_tables', [])
|
333 |
+
if sql_query:
|
334 |
+
is_valid, validation_message = self.validate_sql_query(sql_query, identified_tables)
|
335 |
+
parsed_response['sql_validation'] = {
|
336 |
+
'is_valid': is_valid,
|
337 |
+
'message': validation_message
|
338 |
+
}
|
339 |
+
return {
|
340 |
+
"success": True,
|
341 |
+
"response": parsed_response,
|
342 |
+
"raw_response": response,
|
343 |
+
"model": model_key
|
344 |
+
}
|
345 |
+
except Exception as e:
|
346 |
+
return {
|
347 |
+
"success": False,
|
348 |
+
"error": str(e),
|
349 |
+
"model": model_key
|
350 |
+
}
|
351 |
+
|
352 |
+
def verify_response(self, api_key: str, original_query: str, llama_response: Dict, mistral_response: Dict) -> Dict:
|
353 |
+
"""Use Gemma to verify responses"""
|
354 |
+
self.setup_client(api_key)
|
355 |
+
relevant_schema = self.get_relevant_tables_for_query(original_query)
|
356 |
+
verification_prompt = f"""You are a database query verification expert. You have access to the actual database schema and must verify responses against it.
|
357 |
+
|
358 |
+
ACTUAL DATABASE SCHEMA:
|
359 |
+
{relevant_schema}
|
360 |
+
|
361 |
+
ORIGINAL QUERY: {original_query}
|
362 |
+
|
363 |
+
LLAMA RESPONSE: {json.dumps(llama_response.get('response', {}), indent=2)}
|
364 |
+
|
365 |
+
MISTRAL RESPONSE: {json.dumps(mistral_response.get('response', {}), indent=2)}
|
366 |
+
|
367 |
+
Verify these responses against the ACTUAL schema above. Check:
|
368 |
+
1. Do the table names actually exist in the schema?
|
369 |
+
2. Do the column names actually exist in those tables?
|
370 |
+
3. Are the table selections appropriate for the query?
|
371 |
+
4. Is the SQL syntax correct?
|
372 |
+
5. Are table aliases proper (not reserved words)?
|
373 |
+
|
374 |
+
Respond with ONLY a valid JSON object:
|
375 |
+
{{
|
376 |
+
"verification_summary": "Overall assessment based on actual schema",
|
377 |
+
"table_selection_accuracy": "Assessment of table choices against actual schema",
|
378 |
+
"sql_correctness": "SQL syntax and schema validation",
|
379 |
+
"consistency_check": "Comparison between responses",
|
380 |
+
"recommended_response": "llama, mistral, or neither",
|
381 |
+
"confidence_score": 0.85,
|
382 |
+
"suggested_improvements": ["improvement1", "improvement2"],
|
383 |
+
"potential_issues": ["issue1", "issue2"],
|
384 |
+
"schema_compliance": "Assessment of how well responses match actual schema"
|
385 |
+
}}"""
|
386 |
+
return self.call_model("gemma", verification_prompt, "Verify the above responses against the actual database schema.")
|
387 |
+
|
388 |
+
def execute_query_in_thread(self, sql_query: str, result_queue: queue.Queue):
|
389 |
+
"""Execute SQL query in a thread"""
|
390 |
+
try:
|
391 |
+
if not sql_query.strip().upper().startswith('SELECT'):
|
392 |
+
result_queue.put((False, "Only SELECT queries are allowed"))
|
393 |
+
return
|
394 |
+
sql_query = sql_query.strip().rstrip(';')
|
395 |
+
conn = self.get_db_connection()
|
396 |
+
try:
|
397 |
+
df = pd.read_sql_query(sql_query, conn)
|
398 |
+
result_queue.put((True, df))
|
399 |
+
except Exception as e:
|
400 |
+
result_queue.put((False, str(e)))
|
401 |
+
finally:
|
402 |
+
conn.close()
|
403 |
+
except Exception as e:
|
404 |
+
result_queue.put((False, f"Query execution error: {str(e)}"))
|
405 |
+
|
406 |
+
def execute_query(self, sql_query: str) -> Tuple[bool, Any]:
|
407 |
+
"""Execute SQL query using thread-safe approach"""
|
408 |
+
try:
|
409 |
+
result_queue = queue.Queue()
|
410 |
+
thread = threading.Thread(
|
411 |
+
target=self.execute_query_in_thread,
|
412 |
+
args=(sql_query, result_queue)
|
413 |
+
)
|
414 |
+
thread.start()
|
415 |
+
thread.join(timeout=30)
|
416 |
+
if thread.is_alive():
|
417 |
+
return False, "Query execution timed out"
|
418 |
+
if not result_queue.empty():
|
419 |
+
return result_queue.get()
|
420 |
+
else:
|
421 |
+
return False, "No result returned from query execution"
|
422 |
+
except Exception as e:
|
423 |
+
return False, f"Execution error: {str(e)}"
|
424 |
+
|
425 |
+
def process_query(self, api_key: str, user_query: str) -> Dict:
|
426 |
+
"""Process user query"""
|
427 |
+
if not api_key:
|
428 |
+
return {"error": "Please provide OpenRouter API key"}
|
429 |
+
try:
|
430 |
+
self.setup_client(api_key)
|
431 |
+
system_prompt = self.get_system_prompt(user_query)
|
432 |
+
llama_result = self.call_model("llama", system_prompt, user_query)
|
433 |
+
mistral_result = self.call_model("mistral", system_prompt, user_query)
|
434 |
+
verification_result = self.verify_response(api_key, user_query, llama_result, mistral_result)
|
435 |
+
execution_results = {}
|
436 |
+
for model_name, result in [("llama", llama_result), ("mistral", mistral_result)]:
|
437 |
+
if result.get("success") and result.get("response", {}).get("sql_query"):
|
438 |
+
sql_query = result["response"]["sql_query"]
|
439 |
+
validation_info = result["response"].get("sql_validation", {})
|
440 |
+
if sql_query.strip():
|
441 |
+
if validation_info.get("is_valid", True):
|
442 |
+
success, data = self.execute_query(sql_query)
|
443 |
+
execution_results[model_name] = {
|
444 |
+
"success": success,
|
445 |
+
"data": data.to_dict('records') if success and isinstance(data, pd.DataFrame) else str(data),
|
446 |
+
"row_count": len(data) if success and isinstance(data, pd.DataFrame) else 0,
|
447 |
+
"sql_query": sql_query,
|
448 |
+
"validation": validation_info
|
449 |
+
}
|
450 |
+
else:
|
451 |
+
execution_results[model_name] = {
|
452 |
+
"success": False,
|
453 |
+
"data": f"Query validation failed: {validation_info.get('message', 'Unknown error')}",
|
454 |
+
"row_count": 0,
|
455 |
+
"sql_query": sql_query,
|
456 |
+
"validation": validation_info
|
457 |
+
}
|
458 |
+
else:
|
459 |
+
execution_results[model_name] = {
|
460 |
+
"success": False,
|
461 |
+
"data": "No SQL query generated",
|
462 |
+
"row_count": 0,
|
463 |
+
"sql_query": "",
|
464 |
+
"validation": {"is_valid": False, "message": "Empty query"}
|
465 |
+
}
|
466 |
+
else:
|
467 |
+
execution_results[model_name] = {
|
468 |
+
"success": False,
|
469 |
+
"data": "Model failed to generate response",
|
470 |
+
"row_count": 0,
|
471 |
+
"sql_query": "",
|
472 |
+
"validation": {"is_valid": False, "message": "Model error"}
|
473 |
+
}
|
474 |
+
return {
|
475 |
+
"llama_response": llama_result,
|
476 |
+
"mistral_response": mistral_result,
|
477 |
+
"verification": verification_result,
|
478 |
+
"execution_results": execution_results,
|
479 |
+
"timestamp": datetime.now().isoformat(),
|
480 |
+
"schema_info": self.get_relevant_tables_for_query(user_query)
|
481 |
+
}
|
482 |
+
except Exception as e:
|
483 |
+
return {"error": f"Processing error: {str(e)}", "traceback": traceback.format_exc()}
|
484 |
+
|
485 |
+
def response_to_markdown(response_dict: Dict) -> str:
|
486 |
+
"""Convert model response to Markdown"""
|
487 |
+
if not response_dict.get("success", False):
|
488 |
+
return f"**Error**: {response_dict.get('error', 'Unknown error')}"
|
489 |
+
response = response_dict.get("response", {})
|
490 |
+
markdown = "**Query Analysis Results**\n\n"
|
491 |
+
markdown += f"- **Analysis**: {response.get('analysis', 'N/A')}\n\n"
|
492 |
+
identified_tables = response.get('identified_tables', [])
|
493 |
+
markdown += f"- **Identified Tables**: {', '.join(identified_tables) if identified_tables else 'None'}\n\n"
|
494 |
+
domains_involved = response.get('domains_involved', [])
|
495 |
+
markdown += f"- **Domains Involved**: {', '.join(domains_involved) if domains_involved else 'None'}\n\n"
|
496 |
+
sql_query = response.get('sql_query', '')
|
497 |
+
if sql_query:
|
498 |
+
markdown += "- **SQL Query**:\n\n```sql\n" + sql_query + "\n```\n\n"
|
499 |
+
else:
|
500 |
+
markdown += "- **SQL Query**: None\n\n"
|
501 |
+
markdown += f"- **Explanation**: {response.get('explanation', 'N/A')}\n\n"
|
502 |
+
markdown += f"- **Confidence**: {response.get('confidence', 'N/A')}\n\n"
|
503 |
+
alternative_queries = response.get('alternative_queries', [])
|
504 |
+
if alternative_queries:
|
505 |
+
markdown += "- **Alternative Queries**:\n"
|
506 |
+
for query in alternative_queries:
|
507 |
+
markdown += f" - {query}\n"
|
508 |
+
else:
|
509 |
+
markdown += "- **Alternative Queries**: None\n"
|
510 |
+
validation = response.get('sql_validation', {})
|
511 |
+
if validation:
|
512 |
+
is_valid = validation.get('is_valid', False)
|
513 |
+
message = validation.get('message', 'N/A')
|
514 |
+
markdown += f"\n- **SQL Validation**: {'Passed' if is_valid else 'Failed'} - {message}\n"
|
515 |
+
return markdown
|
516 |
+
|
517 |
+
def verification_to_markdown(verification_dict: Dict) -> str:
|
518 |
+
"""Convert verification response to Markdown"""
|
519 |
+
if not verification_dict.get("success", False):
|
520 |
+
return f"**Error**: {verification_dict.get('error', 'Unknown error')}"
|
521 |
+
response = verification_dict.get("response", {})
|
522 |
+
markdown = "**Verification Results**\n\n"
|
523 |
+
markdown += f"- **Verification Summary**: {response.get('verification_summary', 'N/A')}\n\n"
|
524 |
+
markdown += f"- **Table Selection Accuracy**: {response.get('table_selection_accuracy', 'N/A')}\n\n"
|
525 |
+
markdown += f"- **SQL Correctness**: {response.get('sql_correctness', 'N/A')}\n\n"
|
526 |
+
markdown += f"- **Consistency Check**: {response.get('consistency_check', 'N/A')}\n\n"
|
527 |
+
markdown += f"- **Recommended Response**: {response.get('recommended_response', 'N/A')}\n\n"
|
528 |
+
markdown += f"- **Confidence Score**: {response.get('confidence_score', 'N/A')}\n\n"
|
529 |
+
suggested_improvements = response.get('suggested_improvements', [])
|
530 |
+
if suggested_improvements:
|
531 |
+
markdown += "- **Suggested Improvements**:\n"
|
532 |
+
for improvement in suggested_improvements:
|
533 |
+
markdown += f" - {improvement}\n"
|
534 |
+
else:
|
535 |
+
markdown += "- **Suggested Improvements**: None\n"
|
536 |
+
potential_issues = response.get('potential_issues', [])
|
537 |
+
if potential_issues:
|
538 |
+
markdown += "- **Potential Issues**:\n"
|
539 |
+
for issue in potential_issues:
|
540 |
+
markdown += f" - {issue}\n"
|
541 |
+
else:
|
542 |
+
markdown += "- **Potential Issues**: None\n"
|
543 |
+
markdown += f"- **Schema Compliance**: {response.get('schema_compliance', 'N/A')}\n"
|
544 |
+
return markdown
|
545 |
+
|
546 |
+
def create_gradio_interface():
|
547 |
+
"""Create Gradio interface"""
|
548 |
+
agent = DatabaseQueryAgent()
|
549 |
+
sample_queries = [
|
550 |
+
"Find all customers from customer tables",
|
551 |
+
"Show me employee information from HR tables",
|
552 |
+
"Get patient data from healthcare tables",
|
553 |
+
"List all products with their details",
|
554 |
+
"Find students enrolled in courses",
|
555 |
+
"Show financial transaction records",
|
556 |
+
"Get shipping information for deliveries",
|
557 |
+
"Find all suppliers and their information",
|
558 |
+
"Show retail store data",
|
559 |
+
"Get manufacturing production records"
|
560 |
+
]
|
561 |
+
|
562 |
+
def process_user_query(api_key, query):
|
563 |
+
"""Process query and return formatted results"""
|
564 |
+
if not query.strip():
|
565 |
+
return "Please enter a query", "", "", "", "", ""
|
566 |
+
results = agent.process_query(api_key, query)
|
567 |
+
if "error" in results:
|
568 |
+
return f"**Error**: {results['error']}", "", "", "", "", ""
|
569 |
+
|
570 |
+
# Format responses as Markdown
|
571 |
+
llama_markdown = response_to_markdown(results.get("llama_response", {}))
|
572 |
+
mistral_markdown = response_to_markdown(results.get("mistral_response", {}))
|
573 |
+
verification_markdown = verification_to_markdown(results.get("verification", {}))
|
574 |
+
|
575 |
+
# Format execution results
|
576 |
+
exec_results = results.get("execution_results", {})
|
577 |
+
execution_formatted = ""
|
578 |
+
for model, result in exec_results.items():
|
579 |
+
execution_formatted += f"\n=== {model.upper()} EXECUTION ===\n"
|
580 |
+
execution_formatted += f"SQL Query: {result.get('sql_query', 'N/A')}\n"
|
581 |
+
validation = result.get('validation', {})
|
582 |
+
if validation.get('is_valid'):
|
583 |
+
execution_formatted += f"β
Query Validation: PASSED\n"
|
584 |
+
else:
|
585 |
+
execution_formatted += f"β Query Validation: FAILED - {validation.get('message', 'Unknown error')}\n"
|
586 |
+
if result["success"]:
|
587 |
+
execution_formatted += f"β
Execution: Success! Retrieved {result['row_count']} rows\n"
|
588 |
+
if result["row_count"] > 0:
|
589 |
+
sample_data = result['data'][:3] if isinstance(result['data'], list) else []
|
590 |
+
execution_formatted += f"Sample data:\n{json.dumps(sample_data, indent=2)}\n"
|
591 |
+
else:
|
592 |
+
execution_formatted += "No data returned (empty result set)\n"
|
593 |
+
else:
|
594 |
+
execution_formatted += f"β Execution Error: {result['data']}\n"
|
595 |
+
execution_formatted += "\n"
|
596 |
+
if not execution_formatted:
|
597 |
+
execution_formatted = "No queries were executed. Check if valid SQL was generated."
|
598 |
+
|
599 |
+
schema_info = results.get('schema_info', 'No schema information available')
|
600 |
+
|
601 |
+
# Format summary as Markdown
|
602 |
+
verification_resp = results.get('verification', {}).get('response', {})
|
603 |
+
summary = f"""
|
604 |
+
**π QUERY ANALYSIS COMPLETE**
|
605 |
+
|
606 |
+
ββββββββββββββββββββββββ
|
607 |
+
|
608 |
+
**π Models Used**: Llama 3.1 8B, Mistral 7B, Gemma 2 9B (verification)
|
609 |
+
|
610 |
+
**β° Processed**: {results.get('timestamp', 'N/A')}
|
611 |
+
|
612 |
+
**π― Verification Summary**:
|
613 |
+
|
614 |
+
{verification_resp.get('verification_summary', 'N/A')}
|
615 |
+
|
616 |
+
**π‘ Recommended Model**: {verification_resp.get('recommended_response', 'N/A')}
|
617 |
+
|
618 |
+
**π Confidence**: {verification_resp.get('confidence_score', 'N/A')}
|
619 |
+
|
620 |
+
**ποΈ Schema Compliance**: {verification_resp.get('schema_compliance', 'N/A')}
|
621 |
+
|
622 |
+
**ποΈ Query Execution Status**:
|
623 |
+
|
624 |
+
{len(exec_results)} queries attempted
|
625 |
+
"""
|
626 |
+
|
627 |
+
return summary, llama_markdown, mistral_markdown, verification_markdown, execution_formatted, schema_info
|
628 |
+
|
629 |
+
with gr.Blocks(
|
630 |
+
title="Fixed Intelligent Database Query Agent",
|
631 |
+
theme=gr.themes.Soft(),
|
632 |
+
css="""
|
633 |
+
.gradio-container {
|
634 |
+
max-width: 1200px !important;
|
635 |
+
margin: 0 auto !important;
|
636 |
+
}
|
637 |
+
.result-box {
|
638 |
+
background-color: #f8f9fa;
|
639 |
+
border: 1px solid #dee2e6;
|
640 |
+
border-radius: 8px;
|
641 |
+
padding: 15px;
|
642 |
+
}
|
643 |
+
"""
|
644 |
+
) as interface:
|
645 |
+
gr.HTML("""
|
646 |
+
<div style="text-align: center; padding: 20px; background: linear-gradient(135deg, #667eea 0%, #764ba2 100%); color: white; border-radius: 10px; margin-bottom: 20px;">
|
647 |
+
<h1>π€ Fixed Intelligent Database Query Agent</h1>
|
648 |
+
<p>AI-powered agent that intelligently selects relevant tables from 100+ tables and generates optimized SQL queries</p>
|
649 |
+
<p><strong>Database:</strong> 100 tables across 10 business domains | <strong>Models:</strong> Llama 3.1 8B + Mistral 7B + Gemma 2 9B</p>
|
650 |
+
<p><strong>β
FIXED:</strong> Reserved Word Aliases | Enhanced Column Validation | Better SQL Syntax Checking</p>
|
651 |
+
</div>
|
652 |
+
""")
|
653 |
+
|
654 |
+
with gr.Row():
|
655 |
+
with gr.Column(scale=1):
|
656 |
+
api_key_input = gr.Textbox(
|
657 |
+
label="π OpenRouter API Key",
|
658 |
+
type="password",
|
659 |
+
placeholder="Enter your OpenRouter API key...",
|
660 |
+
info="Get your free API key from openrouter.ai"
|
661 |
+
)
|
662 |
+
query_input = gr.Textbox(
|
663 |
+
label="π¬ Database Query",
|
664 |
+
placeholder="Enter your natural language query...",
|
665 |
+
lines=3,
|
666 |
+
info="Example: 'Find all customers who placed orders in the last month'"
|
667 |
+
)
|
668 |
+
with gr.Row():
|
669 |
+
submit_btn = gr.Button("π Process Query", variant="primary", size="lg")
|
670 |
+
clear_btn = gr.Button("ποΈ Clear", variant="secondary")
|
671 |
+
gr.HTML("<h3>π Sample Test Queries</h3>")
|
672 |
+
sample_dropdown = gr.Dropdown(
|
673 |
+
choices=sample_queries,
|
674 |
+
label="Quick Test Examples",
|
675 |
+
info="Select a sample query to test the agent"
|
676 |
+
)
|
677 |
+
|
678 |
+
with gr.Column(scale=2):
|
679 |
+
summary_output = gr.Markdown(label="π Analysis Summary")
|
680 |
+
with gr.Tabs():
|
681 |
+
with gr.Tab("π¦ Llama 3.1 8B Response"):
|
682 |
+
llama_output = gr.Markdown(label="Llama Response")
|
683 |
+
with gr.Tab("π Mistral 7B Response"):
|
684 |
+
mistral_output = gr.Markdown(label="Mistral Response")
|
685 |
+
with gr.Tab("β
Verification (Gemma 2 9B)"):
|
686 |
+
verification_output = gr.Markdown(label="Verification Analysis")
|
687 |
+
with gr.Tab("ποΈ Query Execution Results"):
|
688 |
+
execution_output = gr.Textbox(
|
689 |
+
label="Database Execution Results",
|
690 |
+
lines=15,
|
691 |
+
max_lines=20,
|
692 |
+
elem_classes=["result-box"]
|
693 |
+
)
|
694 |
+
with gr.Tab("π Database Schema"):
|
695 |
+
schema_output = gr.Textbox(
|
696 |
+
label="Relevant Database Schema",
|
697 |
+
lines=15,
|
698 |
+
max_lines=20,
|
699 |
+
elem_classes=["result-box"]
|
700 |
+
)
|
701 |
+
|
702 |
+
submit_btn.click(
|
703 |
+
fn=process_user_query,
|
704 |
+
inputs=[api_key_input, query_input],
|
705 |
+
outputs=[summary_output, llama_output, mistral_output, verification_output, execution_output, schema_output]
|
706 |
+
)
|
707 |
+
clear_btn.click(
|
708 |
+
fn=lambda: ("", "", "", "", "", "", ""),
|
709 |
+
outputs=[query_input, summary_output, llama_output, mistral_output, verification_output, execution_output, schema_output]
|
710 |
+
)
|
711 |
+
sample_dropdown.change(
|
712 |
+
fn=lambda x: x,
|
713 |
+
inputs=[sample_dropdown],
|
714 |
+
outputs=[query_input]
|
715 |
+
)
|
716 |
+
gr.HTML("""
|
717 |
+
<div style="margin-top: 20px; padding: 15px; background-color: #f8f9fa; border-radius: 8px;">
|
718 |
+
<h3>π― How to Use</h3>
|
719 |
+
<ol>
|
720 |
+
<li><strong>API Key:</strong> Get a free API key from <a href="https://openrouter.ai" target="_blank">openrouter.ai</a></li>
|
721 |
+
<li><strong>Query:</strong> Enter your natural language database query</li>
|
722 |
+
<li><strong>Process:</strong> The agent will analyze your query across 100+ tables and generate optimized SQL</li>
|
723 |
+
<li><strong>Results:</strong> View responses from multiple AI models, verification analysis, and actual query execution results</li>
|
724 |
+
</ol>
|
725 |
+
<p><strong>Features:</strong></p>
|
726 |
+
<ul>
|
727 |
+
<li>π§ Multi-model AI analysis (Llama, Mistral, Gemma)</li>
|
728 |
+
<li>π Intelligent table selection from 100+ tables</li>
|
729 |
+
<li>β
SQL validation and syntax checking</li>
|
730 |
+
<li>ποΈ Real database query execution with results</li>
|
731 |
+
<li>π Cross-model verification and comparison</li>
|
732 |
+
</ul>
|
733 |
+
</div>
|
734 |
+
""")
|
735 |
+
|
736 |
+
return interface
|
737 |
+
|
738 |
+
def main():
|
739 |
+
"""Main function to launch the application"""
|
740 |
+
print("π Starting Intelligent Database Query Agent...")
|
741 |
+
print("π Loading database schema and metadata...")
|
742 |
+
interface = create_gradio_interface()
|
743 |
+
print("β
Database Query Agent Ready!")
|
744 |
+
print("π Access the interface at: http://localhost:7860")
|
745 |
+
print("π Don't forget to add your OpenRouter API key!")
|
746 |
+
interface.launch(share=True)
|
747 |
+
|
748 |
+
if __name__ == "__main__":
|
749 |
main()
|