app.py

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

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

# Load serialized FAISS index and document chunks
chunks, index = joblib.load("rag_model.joblib")

# 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 FAISS (for older FAISS versions)

# Semantic search with fallback handling

# Semantic search using FAISS - strictly for older API with preallocated arrays
def search(query, k=3):
    query_vec = embed_query(query).astype(np.float32)

    # Preallocate arrays (required for FAISS IndexFlatL2 in older versions)
    distances = np.empty((1, k), dtype=np.float32)
    labels = np.empty((1, k), dtype=np.int64)

    # Call FAISS with all required arguments
    index.search(query_vec, k, distances, labels)

    return [chunks[i] for i in labels[0]]

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 = f"Ayudate en el siguiente contexto para responder la pregunta y usa tus conocimientos en caso de ser necesario: \n\n{context}\n\nPregunta: {question}"

    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)