endpoint.py

# endpoint.py
from fastapi import FastAPI, Request
from pydantic import BaseModel
from llama_index.core import Settings, StorageContext, load_index_from_storage
from llama_index.llms.groq import Groq
from llama_index.embeddings.huggingface import HuggingFaceEmbedding
import os,json
from dotenv import load_dotenv

# Load secrets
load_dotenv()
GROQ_API_KEY = os.getenv("GROQ_API_KEY")

# Init LLM and Embedding model
Settings.llm = Groq(model="llama3-8b-8192", api_key=GROQ_API_KEY)
Settings.embed_model = HuggingFaceEmbedding(model_name="sentence-transformers/all-MiniLM-L6-v2")

# Load index
PERSIST_DIR = "./storage"
storage_context = StorageContext.from_defaults(persist_dir=PERSIST_DIR)
index = load_index_from_storage(storage_context)
chat_engine = index.as_chat_engine(chat_mode="context", verbose=False)

app = FastAPI()

class QueryRequest(BaseModel):
    question: str

class RecommendRequest(BaseModel):
    query: str

@app.get("/health")
def health_check():
    return {"status": "healthy"}    

@app.post("/recommend")
async def recommend(request: RecommendRequest):
    prompt = f"""
You are an intelligent assistant that recommends SHL assessments based on user queries.

Using the query: "{request.query}", return **all relevant and matching** SHL assessments (at least 1 and up to 10).

Only respond in this exact JSON format:

{{
  "recommended_assessments": [
    {{
      "url": "Valid URL in string",
      "adaptive_support": "Yes/No",
      "description": "Description in string",
      "duration": 60,
      "remote_support": "Yes/No",
      "test_type": ["List of string"]
    }}
  ]
}}

Do not include any explanations or extra text. Only return pure JSON. Respond with as many matching assessments as possible (up to 10).
"""

    response = chat_engine.chat(prompt)
    try:
        return json.loads(response.response)
    except Exception:
        return {"error": "Model response was not valid JSON", "raw": response.response}