app.py

import streamlit as st
import joblib
import numpy as np
import hnswlib
import os
from openai import OpenAI

# Initialize OpenAI client using secret from Hugging Face Spaces
client = OpenAI(api_key=os.getenv("POCJujitsu"))

# Load serialized HNSW index and document chunks
model_data = joblib.load("rag_model_hnsw.joblib")
chunks = model_data["chunks"]
index = model_data["index"]

# Embed query using OpenAI embedding API
def embed_query(text):
    response = client.embeddings.create(
        model="text-embedding-3-small",
        input=text
    )
    return np.array(response.data[0].embedding, dtype=np.float32).reshape(1, -1)

# Semantic search using HNSWlib
def search(query, k=3):
    query_vec = embed_query(query).astype(np.float32)
    labels, distances = index.knn_query(query_vec, k=k)
    return [chunks[i] for i in labels[0]]

# Chat modes
def chat_no_rag(question):
    response = client.chat.completions.create(
        model="gpt-3.5-turbo",
        messages=[{"role": "user", "content": question}],
        temperature=0.5,
        max_tokens=300
    )
    return response.choices[0].message.content

def chat_with_rag(question, context_chunks):
    context = "\n".join(context_chunks)
    prompt = (
        "Usa el siguiente contexto como referencia para responder la pregunta. "
        "Puedes complementar con tus propios conocimientos si es necesario.\n\n"
        f"Contexto:\n{context}\n\n"
        f"Pregunta: {question}\nRespuesta:"
    )
    response = client.chat.completions.create(
        model="gpt-3.5-turbo",
        messages=[{"role": "user", "content": prompt}],
        temperature=0.3,
        max_tokens=300
    )
    return response.choices[0].message.content

def chat_with_rag_enhanced(question, context_chunks):
    context = "\n".join(context_chunks)
    prompt = (
        "Eres un experto en historia marcial. "
        "Usa el siguiente contexto como referencia para responder la pregunta. "
        "Puedes complementar con tus propios conocimientos si es necesario.\n\n"
        f"Contexto:\n{context}\n\n"
        f"Pregunta: {question}\nRespuesta:"
    )
    response = client.chat.completions.create(
        model="gpt-3.5-turbo",
        messages=[{"role": "user", "content": prompt}],
        temperature=0.2,
        max_tokens=300
    )
    return response.choices[0].message.content

# Streamlit UI
st.set_page_config(page_title="RAG JuJutsu Q&A")
st.title("🤖 JuJutsu AI - Ask Anything")
st.markdown("Ask a question about jujutsu history, techniques, or philosophy.")

question = st.text_input("❓ Enter your question:")
mode = st.radio("Choose response mode:", ["No RAG", "With RAG", "With RAG + Expert Prompt"])

if st.button("Get Answer") and question:
    if mode == "No RAG":
        answer = chat_no_rag(question)
    else:
        retrieved = search(question)
        if mode == "With RAG":
            answer = chat_with_rag(question, retrieved)
        else:
            answer = chat_with_rag_enhanced(question, retrieved)

    st.markdown("### 🧠 Answer")
    st.write(answer)