Cambios para API

app.py

@@ -1,113 +1,91 @@
+
 import streamlit as st
-from PyPDF2 import PdfReader
-from langchain.text_splitter import RecursiveCharacterTextSplitter
-from sentence_transformers import SentenceTransformer
-import openai
-import faiss
+import joblib
 import numpy as np
+import faiss
 import os
-import joblib
-from openai import OpenAI
-
-os.environ["TRANSFORMERS_CACHE"] = "/tmp"
-os.environ["HF_HOME"] = "/tmp"
-os.environ["XDG_CACHE_HOME"] = "/tmp"
-os.environ["SENTENCE_TRANSFORMERS_HOME"] = "/tmp"
-os.environ["STREAMLIT_HOME"] = "/tmp"
-
 from openai import OpenAI
 
+# Initialize OpenAI client using custom environment variable set in Hugging Face
 client = OpenAI(api_key=os.getenv("POCJujitsu"))
 
+# Load serialized FAISS index and document chunks
+chunks, index = joblib.load("rag_model.joblib")
 
-@st.cache_data
-def load_pdf_chunks(pdf_path):
-    reader = PdfReader(pdf_path)
-    raw_text = ""
-    for page in reader.pages:
-        raw_text += page.extract_text() + "\n"
-    splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50)
-    return splitter.split_text(raw_text)
-
-@st.cache_resource
-def load_model_and_index(chunks):
-    model = SentenceTransformer('models/all-MiniLM-L6-v2')
-    embeddings = model.encode(chunks)
-    faiss_index = faiss.IndexFlatL2(embeddings.shape[1])
-    faiss_index.add(np.array(embeddings))
- #   joblib.dump((model, chunks, faiss_index), "rag_model.joblib")
-    return model, chunks, faiss_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)
 
-def search(query, model, chunks, index, k=3):
-    query_vec = model.encode([query])
-    scores, indices = index.search(np.array(query_vec), k)
-    return [chunks[i] for i in indices[0]]
+# Semantic search using FAISS
+def search(query, k=3):
+    query_vec = embed_query(query).astype(np.float32)
+    distances, labels = index.search(query_vec, k)
+    return [chunks[i] for i in labels[0]]
 
-def chat_no_rag(question, max_tokens=250):
+# Chat modes
+def chat_no_rag(question):
     response = client.chat.completions.create(
-        model="gpt-4o",
-        messages=[
-            {"role": "user", "content": question}
-        ],
+        model="gpt-3.5-turbo",
+        messages=[{"role": "user", "content": question}],
         temperature=0.5,
-        max_tokens=max_tokens
+        max_tokens=300
     )
     return response.choices[0].message.content
 
-def chat_with_rag(question, retrieved_chunks, max_tokens=300):
-    context = "\n".join(retrieved_chunks)
-    prompt = f"Usa el siguiente contexto para responder la pregunta:\n\n{context}\n\nPregunta: {question}"
-
+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-4o",
+        model="gpt-3.5-turbo",
         messages=[{"role": "user", "content": prompt}],
         temperature=0.3,
-        max_tokens=max_tokens
+        max_tokens=300
     )
     return response.choices[0].message.content
 
-def chat_with_rag_enhanced(question, retrieved_chunks, max_tokens=300):
-    context = "\n".join(retrieved_chunks)
+def chat_with_rag_enhanced(question, context_chunks):
+    context = "\n".join(context_chunks)
     prompt = (
         "Eres un experto en historia marcial. "
-        "Usa el siguiente contexto histórico para responder con precisión y detalle.\n\n"
+        "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-4o",
+        model="gpt-3.5-turbo",
         messages=[{"role": "user", "content": prompt}],
         temperature=0.2,
-        max_tokens=max_tokens
+        max_tokens=300
     )
     return response.choices[0].message.content
 
 # Streamlit UI
-st.title("📜 RAG JuJutsu Historico - ChatGPT + HF + Streamlit")
-
-if "model" not in st.session_state:
-    with st.spinner("Cargando y procesando el PDF..."):
-        chunks = load_pdf_chunks("JuJutsu-Contexto-Significado-Conexiones-Historia.pdf")
-        model, chunks, index = load_model_and_index(chunks)
-        st.session_state.model = model
-        st.session_state.chunks = chunks
-        st.session_state.index = index
-
-query = st.text_input("Escribe tu pregunta sobre JuJutsu histórico:")
-max_tokens = st.slider("Máximo de tokens de respuesta", 50, 1000, 300, step=50)
-
-if query:
-    model = st.session_state.model
-    chunks = st.session_state.chunks
-    index = st.session_state.index
-
-    st.subheader("🔹 Respuesta sin RAG:")
-    st.write(chat_no_rag(query, max_tokens=max_tokens))
-
-    st.subheader("🔹 Respuesta con RAG:")
-    retrieved = search(query, model, chunks, index)
-    st.write(chat_with_rag(query, retrieved, max_tokens=max_tokens))
-
-    st.subheader("🔹 Respuesta con RAG + Mejora de Prompt:")
-    st.write(chat_with_rag_enhanced(query, retrieved, max_tokens=max_tokens))
+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)