Update app.py

app.py

@@ -1,129 +1,94 @@
 #!/usr/bin/env python3
 """
-RAG Chatbot – Gradio + FAISS + NVIDIA NIM
-Layout com cards automáticos usando examples do ChatInterface
+RAG Chatbot – Hugging Face Space
+Carrega FAISS já existente (sem rebuild)
 """
 
 import os
-import glob
-from typing import List
-
 import gradio as gr
-import pandas as pd
 from openai import OpenAI
 
-from langchain_core.documents import Document
 from langchain_community.vectorstores import FAISS
 from langchain_huggingface import HuggingFaceEmbeddings
-from langchain_text_splitters import RecursiveCharacterTextSplitter
 
-#
+
 # =========================
 # CONFIG
 # =========================
-DATA_DIR = os.getenv("DATA_DIR", "data")
-
+INDEX_DIR = "vectorstore_faiss"
 EMB_MODEL = "sentence-transformers/all-MiniLM-L6-v2"
-CHUNK_SIZE = 900
-CHUNK_OVERLAP = 150
+
 TOP_K = 6
 MAX_CONTEXT_CHARS = 4500
 
-NVIDIA_API_KEY = os.getenv("NVIDIA_API_KEY", "")
+NVIDIA_API_KEY = os.getenv("NVIDIA_API_KEY")
 NVIDIA_BASE_URL = "https://integrate.api.nvidia.com/v1"
 NVIDIA_MODEL = "meta/llama-3.3-70b-instruct"
 
-client = OpenAI(base_url=NVIDIA_BASE_URL, api_key=NVIDIA_API_KEY) if NVIDIA_API_KEY else None
+client = OpenAI(
+    base_url=NVIDIA_BASE_URL,
+    api_key=NVIDIA_API_KEY
+) if NVIDIA_API_KEY else None
 
-SYSTEM_PROMPT = """Você é um assistente que responde perguntas com base em documentos.
-Responda SOMENTE com base no CONTEXTO recuperado.
+
+SYSTEM_PROMPT = """Você responde perguntas usando apenas o CONTEXTO recuperado.
 Se não houver evidência suficiente, diga claramente.
 Seja objetivo.
 """
 
 
 # =========================
-# READ FILES
+# LOAD FAISS EXISTENTE
 # =========================
-SUPPORTED_EXT = {".pdf", ".docx", ".xlsx", ".xls", ".csv", ".txt"}
-
-def list_files(data_dir: str) -> List[str]:
-    files = []
-    for ext in SUPPORTED_EXT:
-        files.extend(glob.glob(os.path.join(data_dir, f"**/*{ext}"), recursive=True))
-    return sorted(set(files))
-
-
-def read_txt(path):
-    try:
-        with open(path, "r", encoding="utf-8", errors="ignore") as f:
-            return f.read()
-    except:
-        return ""
-
-
-def read_csv(path):
-    try:
-        df = pd.read_csv(path)
-        return df.head(1000).to_csv(index=False)
-    except:
-        return ""
-
-
-def read_docx(path):
-    from docx import Document as DocxDocument
-    doc = DocxDocument(path)
-    return "\n".join([p.text for p in doc.paragraphs if p.text.strip()])
-
+embedding = HuggingFaceEmbeddings(model_name=EMB_MODEL)
 
-def read_pdf(path):
-    from pypdf import PdfReader
-    reader = PdfReader(path)
-    return "\n".join([p.extract_text() or "" for p in reader.pages])
+try:
+    vectordb = FAISS.load_local(
+        INDEX_DIR,
+        embedding,
+        allow_dangerous_deserialization=True
+    )
+    STATUS = "✅ Índice FAISS carregado com sucesso."
+except Exception as e:
+    vectordb = None
+    STATUS = f"❌ Erro ao carregar índice: {str(e)}"
 
 
 # =========================
-# BUILD VECTOR DATABASE
+# RAG FUNCTION
 # =========================
-def build_vectordb():
-    files = list_files(DATA_DIR)
-    if not files:
-        raise FileNotFoundError("Nenhum arquivo encontrado na pasta data/")
-
-    docs = []
-    splitter = RecursiveCharacterTextSplitter(
-        chunk_size=CHUNK_SIZE,
-        chunk_overlap=CHUNK_OVERLAP
-    )
+def format_context(docs):
+    context = "\n\n".join([d.page_content for d in docs])
+    if len(context) > MAX_CONTEXT_CHARS:
+        context = context[:MAX_CONTEXT_CHARS]
+    return context
 
-    for path in files:
-        ext = os.path.splitext(path)[1].lower()
-        text = ""
 
-        if ext == ".txt":
-            text = read_txt(path)
-        elif ext == ".csv":
-            text = read_csv(path)
-        elif ext in [".xlsx", ".xls"]:
-            text = read_csv(path)
-        elif ext == ".docx":
-            text = read_docx(path)
-        elif ext == ".pdf":
-            text = read_pdf(path)
+def chat(message, history):
+    if not client:
+        return "❌ Configure NVIDIA_API_KEY em Settings → Secrets."
 
-        for chunk in splitter.split_text(text):
-            docs.append(Document(page_content=chunk, metadata={"source": path}))
+    if not vectordb:
+        return "❌ Índice FAISS não carregado."
 
-    embedding = HuggingFaceEmbeddings(model_name=EMB_MODEL)
-    db = FAISS.from_documents(docs, embedding)
-    return db
+    docs = vectordb.similarity_search(message, k=TOP_K)
+    context = format_context(docs)
 
+    completion = client.chat.completions.create(
+        model=NVIDIA_MODEL,
+        messages=[
+            {"role": "system", "content": SYSTEM_PROMPT},
+            {"role": "user", "content": f"CONTEXTO:\n{context}\n\nPERGUNTA:\n{message}"}
+        ],
+        temperature=0.3,
+        max_tokens=800,
+    )
 
-vectordb = build_vectordb()
+    return completion.choices[0].message.content
 
 
 # =========================
-# SUGGESTIONS (CARDS)
+# SUGGESTION CARDS
 # =========================
 SUGGESTIONS = [
     "Resuma os principais pontos do documento.",
@@ -138,53 +103,23 @@ SUGGESTIONS = [
 
 
 # =========================
-# RAG FUNCTION
+# UI
 # =========================
-def format_context(docs):
-    context = "\n\n".join([d.page_content for d in docs])
-    if len(context) > MAX_CONTEXT_CHARS:
-        context = context[:MAX_CONTEXT_CHARS]
-    return context
-
-
-def chat_rag_nvidia(message, history):
-    if not client:
-        return "❌ Configure NVIDIA_API_KEY."
-
-    retrieved = vectordb.similarity_search(message, k=TOP_K)
-    context = format_context(retrieved)
+with gr.Blocks(title="SOGETREL – RAG Assistant") as demo:
 
-    messages = [
-        {"role": "system", "content": SYSTEM_PROMPT},
-        {"role": "user", "content": f"CONTEXTO:\n{context}\n\nPERGUNTA:\n{message}"}
-    ]
+    gr.Markdown(f"""
+    ## 📚 SOGETREL – Document Assistant
 
-    completion = client.chat.completions.create(
-        model=NVIDIA_MODEL,
-        messages=messages,
-        temperature=0.3,
-        max_tokens=800,
-    )
-
-    return completion.choices[0].message.content
-
-
-# =========================
-# UI (USANDO EXAMPLES NATIVOS)
-# =========================
-with gr.Blocks(title="Document RAG Assistant") as demo:
-
-    gr.Markdown("""
-    ## 📚 SOGETREL
-    Faça perguntas sobre os documentos indexados.
+    {STATUS}
     """)
 
     gr.ChatInterface(
-        fn=chat_rag_nvidia,
-        examples=SUGGESTIONS,  # ← Aqui são gerados os cards automaticamente
+        fn=chat,
+        examples=SUGGESTIONS,
         title="Assistant",
         description="Pergunte algo sobre os documentos."
     )
 
+
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch(server_name="0.0.0.0", server_port=7860)