|
from fastapi import APIRouter, HTTPException, Depends, Query, BackgroundTasks, Request |
|
from typing import List, Optional, Dict, Any |
|
import logging |
|
import time |
|
import os |
|
import json |
|
import hashlib |
|
import asyncio |
|
import traceback |
|
import google.generativeai as genai |
|
from datetime import datetime |
|
from langchain.prompts import PromptTemplate |
|
from langchain_google_genai import GoogleGenerativeAIEmbeddings |
|
from app.utils.utils import timer_decorator |
|
|
|
from app.database.mongodb import get_chat_history, get_request_history, session_collection |
|
from app.database.pinecone import ( |
|
search_vectors, |
|
get_chain, |
|
DEFAULT_TOP_K, |
|
DEFAULT_LIMIT_K, |
|
DEFAULT_SIMILARITY_METRIC, |
|
DEFAULT_SIMILARITY_THRESHOLD, |
|
ALLOWED_METRICS |
|
) |
|
from app.models.rag_models import ( |
|
ChatRequest, |
|
ChatResponse, |
|
ChatResponseInternal, |
|
SourceDocument, |
|
EmbeddingRequest, |
|
EmbeddingResponse, |
|
UserMessageModel |
|
) |
|
|
|
|
|
logger = logging.getLogger(__name__) |
|
|
|
|
|
GOOGLE_API_KEY = os.getenv("GOOGLE_API_KEY") |
|
genai.configure(api_key=GOOGLE_API_KEY) |
|
|
|
|
|
router = APIRouter( |
|
prefix="/rag", |
|
tags=["RAG"], |
|
) |
|
|
|
fix_request = PromptTemplate( |
|
template = """Goal: |
|
Your task is to extract important keywords from the user's current request, optionally using chat history if relevant. |
|
You will receive a conversation history and the user's current message. |
|
Generate a **list of concise keywords** that best represent the user's intent. |
|
|
|
Return Format: |
|
Only return keywords (comma-separated, no extra explanation). |
|
If the current message is NOT related to the chat history or if there is no chat history: Return keywords from the current message only. |
|
If the current message IS related to the chat history: Return a refined set of keywords based on both history and current message. |
|
|
|
Warning: |
|
Only use chat history if the current message is clearly related to the prior context. |
|
|
|
Conversation History: |
|
{chat_history} |
|
|
|
User current message: |
|
{question} |
|
""", |
|
input_variables=["chat_history", "question"], |
|
) |
|
|
|
|
|
prompt = PromptTemplate( |
|
template = """Goal: |
|
You are Pixity - a professional tour guide assistant that assists users in finding information about places in Da Nang, Vietnam. |
|
You can provide details on restaurants, cafes, hotels, attractions, and other local venues. |
|
You have to use core knowledge and conversation history to chat with users, who are Da Nang's tourists. |
|
Pixity’s Core Personality: Friendly & Warm: Chats like a trustworthy friend who listens and is always ready to help. |
|
Naturally Cute: Shows cuteness through word choice, soft emojis, and gentle care for the user. |
|
Playful – a little bit cheeky in a lovable way: Occasionally cracks jokes, uses light memes or throws in a surprise response that makes users smile. Think Duolingo-style humor, but less threatening. |
|
Smart & Proactive: Friendly, but also delivers quick, accurate info. Knows how to guide users to the right place – at the right time – with the right solution. |
|
Tone & Voice: Friendly – Youthful – Snappy. Uses simple words, similar to daily chat language (e.g., “Let’s find it together!” / “Need a tip?” / “Here’s something cool”). Avoids sounding robotic or overly scripted. Can joke lightly in smart ways, making Pixity feel like a travel buddy who knows how to lift the mood |
|
SAMPLE DIALOGUES |
|
When a user opens the chatbot for the first time: |
|
User: Hello? |
|
Pixity: Hi hi 👋 I’ve been waiting for you! Ready to explore Da Nang together? I’ve got tips, tricks, and a tiny bit of magic 🎒✨ |
|
|
|
Return Format: |
|
Respond in friendly, natural, concise and use only English like a real tour guide. |
|
Always use HTML tags (e.g. <b> for bold) so that Telegram can render the special formatting correctly. |
|
|
|
Warning: |
|
Let's support users like a real tour guide, not a bot. The information in core knowledge is your own knowledge. |
|
Your knowledge is provided in the Core Knowledge. All of information in Core Knowledge is about Da Nang, Vietnam. |
|
You just care about current time that user mention when user ask about Solana event. |
|
Only use core knowledge to answer. If you do not have enough information to answer user's question, please reply with "I'm sorry. I don't have information about that" and Give users some more options to ask. |
|
|
|
Core knowledge: |
|
{context} |
|
|
|
Conversation History: |
|
{chat_history} |
|
|
|
User message: |
|
{question} |
|
|
|
Your message: |
|
""", |
|
input_variables = ["context", "question", "chat_history"], |
|
) |
|
|
|
|
|
async def get_embedding(text: str): |
|
"""Get embedding from Google Gemini API""" |
|
try: |
|
|
|
embedding_model = GoogleGenerativeAIEmbeddings(model="models/embedding-001") |
|
|
|
|
|
result = await embedding_model.aembed_query(text) |
|
|
|
|
|
return { |
|
"embedding": result, |
|
"text": text, |
|
"model": "embedding-001" |
|
} |
|
except Exception as e: |
|
logger.error(f"Error generating embedding: {e}") |
|
raise HTTPException(status_code=500, detail=f"Failed to generate embedding: {str(e)}") |
|
|
|
|
|
@router.post("/embedding", response_model=EmbeddingResponse) |
|
async def create_embedding(request: EmbeddingRequest): |
|
""" |
|
Generate embedding for text. |
|
|
|
- **text**: Text to generate embedding for |
|
""" |
|
try: |
|
|
|
embedding_data = await get_embedding(request.text) |
|
|
|
|
|
return EmbeddingResponse(**embedding_data) |
|
except Exception as e: |
|
logger.error(f"Error generating embedding: {e}") |
|
raise HTTPException(status_code=500, detail=f"Failed to generate embedding: {str(e)}") |
|
|
|
@timer_decorator |
|
@router.post("/chat", response_model=ChatResponse) |
|
async def chat(request: ChatRequest, background_tasks: BackgroundTasks): |
|
""" |
|
Get answer for a question using RAG. |
|
|
|
- **user_id**: User's ID from Telegram |
|
- **question**: User's question |
|
- **include_history**: Whether to include user history in prompt (default: True) |
|
- **use_rag**: Whether to use RAG (default: True) |
|
- **similarity_top_k**: Number of top similar documents to return after filtering (default: 6) |
|
- **limit_k**: Maximum number of documents to retrieve from vector store (default: 10) |
|
- **similarity_metric**: Similarity metric to use - cosine, dotproduct, euclidean (default: cosine) |
|
- **similarity_threshold**: Threshold for vector similarity (default: 0.75) |
|
- **session_id**: Optional session ID for tracking conversations |
|
- **first_name**: User's first name |
|
- **last_name**: User's last name |
|
- **username**: User's username |
|
""" |
|
start_time = time.time() |
|
try: |
|
|
|
session_id = request.session_id or f"{request.user_id}_{datetime.now().strftime('%Y-%m-%d_%H:%M:%S')}" |
|
|
|
|
|
retriever = get_chain( |
|
top_k=request.similarity_top_k, |
|
limit_k=request.limit_k, |
|
similarity_metric=request.similarity_metric, |
|
similarity_threshold=request.similarity_threshold |
|
) |
|
if not retriever: |
|
raise HTTPException(status_code=500, detail="Failed to initialize retriever") |
|
|
|
|
|
chat_history = get_chat_history(request.user_id) if request.include_history else "" |
|
logger.info(f"Using chat history: {chat_history[:100]}...") |
|
|
|
|
|
generation_config = { |
|
"temperature": 0.9, |
|
"top_p": 1, |
|
"top_k": 1, |
|
"max_output_tokens": 2048, |
|
} |
|
|
|
safety_settings = [ |
|
{ |
|
"category": "HARM_CATEGORY_HARASSMENT", |
|
"threshold": "BLOCK_MEDIUM_AND_ABOVE" |
|
}, |
|
{ |
|
"category": "HARM_CATEGORY_HATE_SPEECH", |
|
"threshold": "BLOCK_MEDIUM_AND_ABOVE" |
|
}, |
|
{ |
|
"category": "HARM_CATEGORY_SEXUALLY_EXPLICIT", |
|
"threshold": "BLOCK_MEDIUM_AND_ABOVE" |
|
}, |
|
{ |
|
"category": "HARM_CATEGORY_DANGEROUS_CONTENT", |
|
"threshold": "BLOCK_MEDIUM_AND_ABOVE" |
|
}, |
|
] |
|
|
|
model = genai.GenerativeModel( |
|
model_name='models/gemini-2.0-flash', |
|
generation_config=generation_config, |
|
safety_settings=safety_settings |
|
) |
|
|
|
prompt_request = fix_request.format( |
|
question=request.question, |
|
chat_history=chat_history |
|
) |
|
|
|
|
|
final_request_start_time = time.time() |
|
final_request = model.generate_content(prompt_request) |
|
|
|
logger.info(f"Fixed Request: {final_request.text}") |
|
logger.info(f"Final request generation time: {time.time() - final_request_start_time:.2f} seconds") |
|
|
|
|
|
retrieved_docs = retriever.invoke(final_request.text) |
|
logger.info(f"Retrieve: {retrieved_docs}") |
|
context = "\n".join([doc.page_content for doc in retrieved_docs]) |
|
|
|
sources = [] |
|
for doc in retrieved_docs: |
|
source = None |
|
metadata = {} |
|
|
|
if hasattr(doc, 'metadata'): |
|
source = doc.metadata.get('source', None) |
|
|
|
score = doc.metadata.get('score', None) |
|
normalized_score = doc.metadata.get('normalized_score', None) |
|
|
|
metadata = {k: v for k, v in doc.metadata.items() |
|
if k not in ['text', 'source', 'score', 'normalized_score']} |
|
|
|
sources.append(SourceDocument( |
|
text=doc.page_content, |
|
source=source, |
|
score=score, |
|
normalized_score=normalized_score, |
|
metadata=metadata |
|
)) |
|
|
|
|
|
prompt_text = prompt.format( |
|
context=context, |
|
question=request.question, |
|
chat_history=chat_history |
|
) |
|
logger.info(f"Full prompt with history and context: {prompt_text}") |
|
|
|
|
|
response = model.generate_content(prompt_text) |
|
answer = response.text |
|
|
|
|
|
processing_time = time.time() - start_time |
|
|
|
|
|
|
|
|
|
|
|
chat_response = ChatResponse( |
|
answer=answer, |
|
processing_time=processing_time |
|
) |
|
|
|
|
|
return chat_response |
|
except Exception as e: |
|
logger.error(f"Error processing chat request: {e}") |
|
import traceback |
|
logger.error(traceback.format_exc()) |
|
raise HTTPException(status_code=500, detail=f"Failed to process chat request: {str(e)}") |
|
|
|
|
|
@router.get("/health") |
|
async def health_check(): |
|
""" |
|
Check health of RAG services and retrieval system. |
|
|
|
Returns: |
|
- status: "healthy" if all services are working, "degraded" otherwise |
|
- services: Status of each service (gemini, pinecone) |
|
- retrieval_config: Current retrieval configuration |
|
- timestamp: Current time |
|
""" |
|
services = { |
|
"gemini": False, |
|
"pinecone": False |
|
} |
|
|
|
|
|
try: |
|
|
|
model = genai.GenerativeModel("gemini-2.0-flash") |
|
|
|
response = model.generate_content("Hello") |
|
services["gemini"] = True |
|
except Exception as e: |
|
logger.error(f"Gemini health check failed: {e}") |
|
|
|
|
|
try: |
|
|
|
from app.database.pinecone import get_pinecone_index |
|
|
|
index = get_pinecone_index() |
|
|
|
if index: |
|
services["pinecone"] = True |
|
except Exception as e: |
|
logger.error(f"Pinecone health check failed: {e}") |
|
|
|
|
|
retrieval_config = { |
|
"default_top_k": DEFAULT_TOP_K, |
|
"default_limit_k": DEFAULT_LIMIT_K, |
|
"default_similarity_metric": DEFAULT_SIMILARITY_METRIC, |
|
"default_similarity_threshold": DEFAULT_SIMILARITY_THRESHOLD, |
|
"allowed_metrics": ALLOWED_METRICS |
|
} |
|
|
|
|
|
status = "healthy" if all(services.values()) else "degraded" |
|
return { |
|
"status": status, |
|
"services": services, |
|
"retrieval_config": retrieval_config, |
|
"timestamp": datetime.now().isoformat() |
|
} |