rag_module.py

# rag_module.py
import json
from typing import Dict
from embedder import embed_text
from vectorstore_local import query_similar
from ai_module import get_ai_response

def _load_config():
    with open("config.json", "r", encoding="utf-8") as f:
        return json.load(f)

def build_prompt(question: str, passages):
    context_block = "\n\n".join(
        [f"[Source: {p['meta'].get('source','unknown')}]\n{p['text']}" for p in passages]
    )
    instructions = (
        "You are Jarvis, a helpful assistant. Answer the user's question using ONLY the provided context. "
        "If the answer is not in the context, say you don't know. Keep it concise.\n"
    )
    return f"{instructions}\nContext:\n{context_block}\n\nQuestion: {question}\nAnswer:"

def ask_knowledge(question: str) -> str:
    cfg = _load_config()
    top_k = cfg["rag"]["top_k"]
    q_emb = embed_text(question)
    passages = query_similar(q_emb, top_k)
    if not passages:
        return "I couldn't find anything relevant in your knowledge base."
    prompt = build_prompt(question, passages)
    return get_ai_response(prompt)