Naz786's picture
Create app.py
8b44e94 verified
raw
history blame
14.9 kB
import streamlit as st
import difflib
import os
import re
import hashlib
from groq import Groq
# --- Page config ---
st.set_page_config(page_title="πŸš€ AI Assistant with Workflow + Semantic Search", layout="wide")
# --- Groq API Setup ---
GROQ_API_KEY = os.environ.get("GROQ_API_KEY")
if not GROQ_API_KEY:
st.error("❌ Please set your GROQ_API_KEY environment variable.")
st.stop()
client = Groq(api_key=GROQ_API_KEY)
# --- Cache for embeddings ---
embedding_cache = {}
def get_embedding(text):
key = hashlib.sha256(text.encode()).hexdigest()
if key in embedding_cache:
return embedding_cache[key]
embedding = [ord(c) % 100 / 100 for c in text[:512]]
embedding_cache[key] = embedding
return embedding
def cosine_similarity(vec1, vec2):
dot = sum(a*b for a,b in zip(vec1, vec2))
norm1 = sum(a*a for a in vec1) ** 0.5
norm2 = sum(b*b for b in vec2) ** 0.5
return dot / (norm1 * norm2 + 1e-8)
def split_code_into_chunks(code, lang):
if lang.lower() == "python":
pattern = r'(def\s+\w+\(.*?\):|class\s+\w+\(?.*?\)?:)'
splits = re.split(pattern, code)
chunks = []
for i in range(1, len(splits), 2):
header = splits[i]
body = splits[i+1] if (i+1) < len(splits) else ""
chunks.append(header + body)
return chunks if chunks else [code]
else:
return [code]
def groq_call(prompt):
resp = client.chat.completions.create(
messages=[{"role": "user", "content": prompt}],
model="llama3-70b-8192",
)
return resp.choices[0].message.content
def semantic_search_improved(code, question, lang, skill, role, explain_lang):
chunks = split_code_into_chunks(code, lang)
question_emb = get_embedding(question)
scored_chunks = []
for chunk in chunks:
emb = get_embedding(chunk)
score = cosine_similarity(question_emb, emb)
scored_chunks.append((score, chunk))
scored_chunks.sort(key=lambda x: x[0], reverse=True)
top_chunks = [c for _, c in scored_chunks[:3]]
combined_code = "\n\n".join(top_chunks)
prompt = (
f"You are a friendly and insightful {lang} expert helping a {skill} {role}.\n"
f"Based on these relevant code snippets:\n{combined_code}\n"
f"Answer this question in {explain_lang}:\n{question}\n"
f"Explain which parts handle the question and how to modify them if needed."
)
return groq_call(prompt)
def error_detection_and_fixes(refactored_code, lang, skill, role, explain_lang):
prompt = (
f"You are a senior {lang} developer. Analyze this code for bugs, security flaws, "
f"and performance issues. Suggest fixes with explanations in {explain_lang}:\n\n{refactored_code}"
)
return groq_call(prompt)
def agentic_workflow(code, skill_level, programming_language, explanation_language, user_role):
timeline = []
suggestions = []
# Explanation
explain_prompt = (
f"You are a friendly and insightful {programming_language} expert helping a {skill_level} {user_role}. "
f"Explain this code in {explanation_language} with clear examples, analogies, and why each part matters:\n\n{code}"
)
explanation = groq_call(explain_prompt)
timeline.append({"step": "Explain", "description": "Detailed explanation", "output": explanation, "code": code})
suggestions.append("Consider refactoring your code to improve readability and performance.")
# Refactor
refactor_prompt = (
f"Refactor this {programming_language} code. Explain the changes like a mentor helping a {skill_level} {user_role}. "
f"Include best practices and improvements:\n\n{code}"
)
refactor_response = groq_call(refactor_prompt)
if "```" in refactor_response:
parts = refactor_response.split("```")
refactored_code = ""
for part in parts:
if part.strip().startswith(programming_language.lower()):
refactored_code = part.strip().split('\n', 1)[1] if '\n' in part else ""
break
if not refactored_code:
refactored_code = refactor_response
else:
refactored_code = refactor_response
timeline.append({"step": "Refactor", "description": "Refactored code with improvements", "output": refactored_code, "code": refactored_code})
suggestions.append("Review the refactored code and adapt it to your style or project needs.")
# Review
review_prompt = (
f"As a senior {programming_language} developer, review the refactored code. "
f"Give constructive feedback on strengths, weaknesses, performance, security, and improvements in {explanation_language}:\n\n{refactored_code}"
)
review = groq_call(review_prompt)
timeline.append({"step": "Review", "description": "Code review and suggestions", "output": review, "code": refactored_code})
suggestions.append("Incorporate review feedback for cleaner, robust code.")
# Error detection & fixes
errors = error_detection_and_fixes(refactored_code, programming_language, skill_level, user_role, explanation_language)
timeline.append({"step": "Error Detection", "description": "Bugs, security, performance suggestions", "output": errors, "code": refactored_code})
suggestions.append("Apply fixes to improve code safety and performance.")
# Test generation
test_prompt = (
f"Write clear, effective unit tests for this {programming_language} code. "
f"Explain what each test does in {explanation_language}, for a {skill_level} {user_role}:\n\n{refactored_code}"
)
tests = groq_call(test_prompt)
timeline.append({"step": "Test Generation", "description": "Generated unit tests", "output": tests, "code": tests})
suggestions.append("Run generated tests locally to validate changes.")
return timeline, suggestions
def get_inline_diff_html(original, modified):
differ = difflib.HtmlDiff(tabsize=4, wrapcolumn=80)
html = differ.make_table(
original.splitlines(), modified.splitlines(),
"Original", "Refactored", context=True, numlines=2
)
return f'<div style="overflow-x:auto; max-height:400px;">{html}</div>'
def detect_code_type(code, programming_language):
backend_keywords = [
'flask', 'django', 'express', 'fastapi', 'spring', 'controller', 'api', 'server', 'database', 'sql', 'mongoose'
]
frontend_keywords = [
'react', 'vue', 'angular', 'component', 'html', 'css', 'document.getelementbyid', 'window.', 'render', 'jsx',
'<html', '<body', '<script', '<div', 'getelementbyid', 'queryselector', 'addeventlistener', 'innerhtml'
]
data_science_keywords = [
'pandas', 'numpy', 'sklearn', 'matplotlib', 'seaborn', 'plt', 'train_test_split', 'randomforestclassifier', 'classification_report'
]
code_lower = code.lower()
if any(word in code_lower for word in data_science_keywords):
return 'data_science'
if any(word in code_lower for word in frontend_keywords):
return 'frontend'
if programming_language.lower() in ['python', 'java', 'c#']:
if any(word in code_lower for word in backend_keywords):
return 'backend'
if programming_language.lower() in ['javascript', 'typescript', 'java', 'c#']:
if any(word in code_lower for word in frontend_keywords):
return 'frontend'
if programming_language.lower() in ['python', 'java', 'c#']:
return 'backend'
if programming_language.lower() in ['javascript', 'typescript']:
return 'frontend'
return 'unknown'
def code_complexity(code):
lines = code.count('\n') + 1
functions = code.count('def ')
classes = code.count('class ')
comments = code.count('#')
return f"Lines: {lines}, Functions: {functions}, Classes: {classes}, Comments: {comments}"
def code_matches_language(code: str, language: str) -> bool:
"""Strictly check whether code matches key patterns of the selected language."""
code_lower = code.strip().lower()
language = language.lower()
patterns = {
"python": [
"def ", "class ", "import ", "from ", "try:", "except", "raise", "lambda",
"with ", "yield", "async ", "await", "print(", "self.", "__init__", "__name__",
"if __name__ == '__main__':", "#!", # shebang for executable scripts
],
"c++": [
"#include", "int main(", "std::", "::", "cout <<", "cin >>", "new ", "delete ",
"try {", "catch(", "template<", "using namespace", "class ", "struct ", "#define",
],
"java": [
"package ", "import java.", "public class", "private ", "protected ", "public static void main",
"System.out.println", "try {", "catch(", "throw new ", "implements ", "extends ",
"@Override", "interface ", "enum ", "synchronized ", "final ",
],
"c#": [
"using System", "namespace ", "class ", "interface ", "public static void Main",
"Console.WriteLine", "try {", "catch(", "throw ", "async ", "await ", "get;", "set;",
"List<", "Dictionary<", "[Serializable]", "[Obsolete]",
],
"javascript": [
"function ", "const ", "let ", "var ", "document.", "window.", "console.log",
"if(", "for(", "while(", "switch(", "try {", "catch(", "export ", "import ", "async ",
"await ", "=>", "this.", "class ", "prototype", "new ", "$(",
],
"typescript": [
"function ", "const ", "let ", "interface ", "type ", ": string", ": number", ": boolean",
"implements ", "extends ", "enum ", "public ", "private ", "protected ", "readonly ",
"import ", "export ", "console.log", "async ", "await ", "=>", "this.",
],
"html": [
"<!doctype html", "<html", "<head>", "<body>", "<script", "<style", "<meta ", "<link ",
"<title>", "<div", "<span", "<p>", "<h1>", "<ul>", "<li>", "<form", "<input", "<button",
"<table", "<footer", "<header", "<section", "<article", "<nav", "<img", "<a ", "</html>",
],
}
match_patterns = patterns.get(language, [])
match_count = sum(1 for pattern in match_patterns if pattern in code_lower)
# Require at least one pattern to match for validation to succeed
return match_count >= 1
# --- Sidebar ---
st.sidebar.title("πŸ”§ Configuration")
lang = st.sidebar.selectbox("Programming Language", ["Python", "JavaScript", "C++", "Java", "C#", "TypeScript"])
skill = st.sidebar.selectbox("Skill Level", ["Beginner", "Intermediate", "Expert"])
role = st.sidebar.selectbox("Your Role", ["Student", "Frontend Developer", "Backend Developer", "Data Scientist"])
explain_lang = st.sidebar.selectbox("Explanation Language", ["English", "Spanish", "Chinese", "Urdu"])
st.sidebar.markdown("---")
st.sidebar.markdown("<span style='color:#fff;'>Powered by <b>BLACKBOX.AI</b></span>", unsafe_allow_html=True)
tabs = st.tabs(["🧠 Full AI Workflow", "πŸ” Semantic Search"])
# --- Tab 1: Full AI Workflow ---
with tabs[0]:
st.title("🧠 Full AI Workflow")
file_types = {
"Python": ["py"],
"JavaScript": ["js"],
"C++": ["cpp", "h", "hpp"],
"Java": ["java"],
"C#": ["cs"],
"TypeScript": ["ts"],
}
uploaded_file = st.file_uploader(
f"Upload {', '.join(file_types.get(lang, []))} file(s)",
type=file_types.get(lang, None)
)
if uploaded_file:
code_input = uploaded_file.read().decode("utf-8")
else:
code_input = st.text_area("Your Code", height=300, placeholder="Paste your code here...")
if code_input:
st.markdown(f"<b>Complexity:</b> {code_complexity(code_input)}", unsafe_allow_html=True)
if st.button("Run AI Workflow"):
if not code_input.strip():
st.warning("Please paste or upload your code.")
elif not code_matches_language(code_input, lang):
st.error(f"The pasted code doesn’t look like valid {lang} code. Please check your code or select the correct language.")
else:
code_type = detect_code_type(code_input, lang)
if code_type == "data_science" and role != "Data Scientist":
st.error("Data science code detected. Please select 'Data Scientist' role.")
elif code_type == "frontend" and role != "Frontend Developer":
st.error("Frontend code detected. Please select 'Frontend Developer' role.")
elif code_type == "backend" and role != "Backend Developer":
st.error("Backend code detected. Please select 'Backend Developer' role.")
else:
with st.spinner("Running agentic workflow..."):
timeline, suggestions = agentic_workflow(code_input, skill, lang, explain_lang, role)
# Show each step in an expander
for step in timeline:
with st.expander(f"βœ… {step['step']} - {step['description']}"):
if step['step'] == "Refactor":
diff_html = get_inline_diff_html(code_input, step['code'])
st.markdown(diff_html, unsafe_allow_html=True)
st.code(step['output'], language=lang.lower())
else:
st.markdown(step['output'])
st.markdown("#### Agent Suggestions")
for s in suggestions:
st.markdown(f"- {s}")
# Download buttons after suggestions
st.markdown("---")
st.markdown("### πŸ“₯ Download Results")
report_text = ""
for step in timeline:
report_text += f"## {step['step']}\n{step['description']}\n\n{step['output']}\n\n"
st.download_button(
label="πŸ“„ Download Full Workflow Report",
data=report_text,
file_name="ai_workflow_report.txt",
mime="text/plain",
)
# --- Tab 2: Semantic Search ---
with tabs[1]:
st.title("πŸ” Semantic Search")
sem_code = st.text_area("Your Code", height=300, placeholder="Paste your code...")
sem_q = st.text_input("Your Question", placeholder="E.g., What does this function do?")
if st.button("Run Semantic Search"):
if not sem_code.strip() or not sem_q.strip():
st.warning("Code and question required.")
else:
with st.spinner("Running semantic search..."):
answer = semantic_search_improved(sem_code, sem_q, lang, skill, role, explain_lang)
st.markdown("### πŸ“Œ Answer")
st.markdown(answer)
st.markdown("---")