app.py

from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
from utils import retrive_context, generate_response


# Initialize FastAPI
app = FastAPI()


class QueryRequest(BaseModel):
    # Asked query should be in string format
    query: str             


class QueryResponse(BaseModel):
    # Response should be in string format
    response: str



@app.post("/infer", response_model=QueryResponse)
def infer(query_request: QueryRequest):
    query = query_request.query
    context = retrive_context(query)
    if context == 500:
        raise HTTPException(status_code=500, detail="Error retrieving context")
    
    response = generate_response(query, context)
    if response == 500:
        raise HTTPException(status_code=500, detail="Error generating response")
    
    return QueryResponse(response=response)


# Root endpoint for testing
@app.get("/")
def read_root():
    return {"message": "Inference API is running"}

if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="127.0.0.1", port=8000, log_level="info")