app.py

"""

LexiBot FastAPI Backend

Headless RAG API for Indian Legal Information



Replaces the legacy Telegram bot (main.py) with a REST API.

Designed for deployment on Hugging Face Spaces.

"""

import os
from typing import Dict
from contextlib import asynccontextmanager
from dotenv import load_dotenv

from fastapi import FastAPI, HTTPException
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel

from langchain_google_genai import ChatGoogleGenerativeAI
from langchain_pinecone import PineconeEmbeddings, PineconeVectorStore
from langchain.chains import RetrievalQA
from langchain.memory import ConversationBufferWindowMemory
from langchain.prompts import PromptTemplate

load_dotenv()

# Configuration
GOOGLE_API_KEY = os.getenv("GOOGLE_API_KEY")
PINECONE_API_KEY = os.getenv("PINECONE_API_KEY")
PINECONE_INDEX_NAME = os.getenv("PINECONE_INDEX_NAME")

# Global state
vector_store = None
llm = None
embeddings = None
session_memories: Dict[str, ConversationBufferWindowMemory] = {}


# Pydantic Models
class ChatRequest(BaseModel):
    message: str
    session_id: str


class ChatResponse(BaseModel):
    response: str
    sources: list[str]


class HealthResponse(BaseModel):
    status: str


# Legal RAG Prompt
LEGAL_PROMPT = PromptTemplate(
    template="""You are LexiBot, an AI legal assistant specializing in Indian law.



IMPORTANT GUIDELINES:

- Provide accurate information based ONLY on the context provided

- If the context doesn't contain relevant information, say "I don't have specific information about that in my legal database"

- Always recommend consulting a qualified lawyer for specific legal matters

- Be clear, concise, and use simple language

- When citing laws, mention the specific Act name and section number



CONTEXT FROM LEGAL DATABASE:

{context}



USER QUESTION: {question}



Provide a helpful, accurate response based on the legal context above:""",
    input_variables=["context", "question"]
)


@asynccontextmanager
async def lifespan(app: FastAPI):
    """Initialize resources on startup."""
    global vector_store, llm, embeddings
    
    print("🚀 Initializing LexiBot API...")
    
    # Validate environment
    if not GOOGLE_API_KEY:
        raise ValueError("GOOGLE_API_KEY not set")
    if not PINECONE_API_KEY:
        raise ValueError("PINECONE_API_KEY not set")
    
    # Initialize Pinecone embeddings (same as ingestion script)
    print("   Loading Pinecone Embeddings (multilingual-e5-large)...")
    embeddings = PineconeEmbeddings(
        model="multilingual-e5-large",
        pinecone_api_key=PINECONE_API_KEY
    )
    
    # Initialize LLM
    print("   Loading Gemini LLM...")
    llm = ChatGoogleGenerativeAI(
        model="gemini-2.5-flash",
        google_api_key=GOOGLE_API_KEY,
        temperature=0.1,
        max_tokens=2048
    )
    
    # Connect to Pinecone
    print("   Connecting to Pinecone...")
    vector_store = PineconeVectorStore(
        index_name=PINECONE_INDEX_NAME,
        embedding=embeddings,
        pinecone_api_key=PINECONE_API_KEY
    )
    
    print("✅ LexiBot API Ready!")
    
    yield
    
    # Cleanup
    print("👋 Shutting down LexiBot API...")


# Initialize FastAPI
app = FastAPI(
    title="LexiBot API",
    description="Headless RAG API for Indian Legal Information",
    version="2.0.0",
    lifespan=lifespan
)

# CORS Middleware - Allow all origins for frontend integration
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],  # Configure appropriately for production
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)


def get_session_memory(session_id: str) -> ConversationBufferWindowMemory:
    """

    Get or create conversation memory for a session.

    

    Note: Memory is ephemeral (RAM-based). If the container restarts, memory clears.

    This is acceptable for the MVP as per PRD requirements.

    """
    if session_id not in session_memories:
        session_memories[session_id] = ConversationBufferWindowMemory(
            k=5,  # Keep last 5 exchanges
            memory_key="chat_history",
            return_messages=True
        )
    return session_memories[session_id]


def extract_sources(source_documents) -> list[str]:
    """Extract unique act names from source documents."""
    sources = set()
    for doc in source_documents:
        if "act_name" in doc.metadata:
            sources.add(doc.metadata["act_name"])
        elif "source" in doc.metadata:
            sources.add(doc.metadata["source"])
    return list(sources)


@app.get("/", response_model=HealthResponse)
async def root():
    """Root endpoint - health check."""
    return HealthResponse(status="ok")


@app.get("/health", response_model=HealthResponse)
async def health_check():
    """Health check endpoint for uptime monitoring."""
    return HealthResponse(status="ok")


@app.post("/chat", response_model=ChatResponse)
async def chat(request: ChatRequest):
    """

    Main chat endpoint for legal queries.

    

    - **message**: The user's legal question

    - **session_id**: Unique session identifier for conversation memory

    

    Returns:

    - **response**: The AI-generated legal response

    - **sources**: List of legal acts referenced in the response

    """
    if not vector_store or not llm:
        raise HTTPException(status_code=503, detail="Service not initialized")
    
    if not request.message.strip():
        raise HTTPException(status_code=400, detail="Message cannot be empty")
    
    try:
        # Get session memory
        memory = get_session_memory(request.session_id)
        
        # Create QA chain
        qa_chain = RetrievalQA.from_chain_type(
            llm=llm,
            chain_type="stuff",
            retriever=vector_store.as_retriever(search_kwargs={"k": 5}),
            return_source_documents=True,
            chain_type_kwargs={"prompt": LEGAL_PROMPT}
        )
        
        # Execute query
        result = qa_chain.invoke({"query": request.message})
        
        # Extract response and sources
        response_text = result.get("result", "I couldn't process your query.")
        source_docs = result.get("source_documents", [])
        sources = extract_sources(source_docs)
        
        # Save to memory
        memory.save_context(
            {"input": request.message},
            {"output": response_text}
        )
        
        return ChatResponse(
            response=response_text,
            sources=sources
        )
        
    except Exception as e:
        print(f"Error processing chat: {e}")
        raise HTTPException(status_code=500, detail=f"Error processing request: {str(e)}")


if __name__ == "__main__":
    import uvicorn
    uvicorn.run("app:app", host="0.0.0.0", port=7860, reload=True)