app.py

import gradio as gr
import os
from langchain_community.document_loaders import PyPDFDirectoryLoader
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_huggingface import HuggingFaceEmbeddings, HuggingFaceEndpoint
from langchain_community.vectorstores import FAISS
from langchain.chains import create_retrieval_chain
from langchain.chains.combine_documents import create_stuff_documents_chain
from langchain_core.prompts import ChatPromptTemplate

DATA_DIR = "Data" if os.path.exists("Data") else "data"
if not os.path.exists(DATA_DIR):
    os.makedirs(DATA_DIR)

print(f"Usando diretorio: {DATA_DIR}")
print(f"Arquivos: {os.listdir(DATA_DIR) if os.path.exists(DATA_DIR) else 'vazio'}")

loader = PyPDFDirectoryLoader(DATA_DIR)
documents = loader.load()
print(f"Documentos carregados: {len(documents)}")

text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=150)
docs = text_splitter.split_documents(documents)

model_name = "sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2"
embeddings = HuggingFaceEmbeddings(model_name=model_name)

if docs:
    db = FAISS.from_documents(docs, embeddings)
else:
    import faiss
    dimension = 384
    index = faiss.IndexFlatL2(dimension)
    from langchain_community.docstore.in_memory import InMemoryDocstore
    db = FAISS(embedding_function=embeddings, index=index, docstore=InMemoryDocstore(), index_to_docstore_id={})

retriever = db.as_retriever(search_kwargs={"k": 3})

# Usando HuggingFaceEndpoint (substituto moderno do HuggingFaceHub)
llm = HuggingFaceEndpoint(
    repo_id="mistralai/Mistral-7B-Instruct-v0.3",
    task="text-generation",
    huggingfacehub_api_token=os.getenv("HUGGINGFACEHUB_API_TOKEN"),
    max_new_tokens=512,
    temperature=0.7
)

system_prompt = (
    "Você é o Prometheus, um assistente especializado no DETRAN-RJ. "
    "Use o contexto abaixo para responder às perguntas dos usuários de forma clara e profissional. "
    "Se não souber a resposta com base no contexto, diga que não encontrou a informação específica, "
    "mas tente ajudar com o que for possível.\n\n"
    "{context}"
)

prompt = ChatPromptTemplate.from_messages([
    ("system", system_prompt),
    ("human", "{input}"),
])

question_answer_chain = create_stuff_documents_chain(llm, prompt)
rag_chain = create_retrieval_chain(retriever, question_answer_chain)

def respond(message, history):
    try:
        response = rag_chain.invoke({"input": message})
        return response["answer"]
    except Exception as e:
        return f"Erro ao processar consulta: {str(e)}"

demo = gr.ChatInterface(
    respond,
    title="Prometheus — Agente Especialista DETRAN-RJ",
    description="IA treinada com regulamentações e manuais do DETRAN-RJ para suporte ao cidadão.",
    examples=["Como renovar a CNH?", "Quais os documentos para transferência de veículo?", "O que é o GRT?"]
)

if __name__ == "__main__":
    demo.launch()
    import gradio as gr
import os
from langchain_community.document_loaders import PyPDFDirectoryLoader
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_community.vectorstores import FAISS
from langchain.chains import create_retrieval_chain
from langchain.chains.combine_documents import create_stuff_documents_chain
from langchain_core.prompts import ChatPromptTemplate
from langchain_community.llms import HuggingFaceHub

DATA_DIR = "Data" if os.path.exists("Data") else "data"
if not os.path.exists(DATA_DIR):
    os.makedirs(DATA_DIR)

print(f"Usando diretorio: {DATA_DIR}")
print(f"Arquivos: {os.listdir(DATA_DIR) if os.path.exists(DATA_DIR) else 'vazio'}")

loader = PyPDFDirectoryLoader(DATA_DIR)
documents = loader.load()
print(f"Documentos carregados: {len(documents)}")

text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=150)
docs = text_splitter.split_documents(documents)

model_name = "sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2"
embeddings = HuggingFaceEmbeddings(model_name=model_name)

if docs:
    db = FAISS.from_documents(docs, embeddings)
else:
    import faiss
    dimension = 384
    index = faiss.IndexFlatL2(dimension)
    from langchain_community.docstore.in_memory import InMemoryDocstore
    db = FAISS(embedding_function=embeddings, index=index,
               docstore=InMemoryDocstore({}), index_to_docstore_id={})

repo_id = "mistralai/Mixtral-8x7B-Instruct-v0.1"
llm = HuggingFaceHub(
    repo_id=repo_id,
    task="text-generation",
    model_kwargs={"temperature": 0.2, "max_new_tokens": 1024}
)

system_prompt = (
    "Voce e um assistente especializado em analise de contratos do DETRAN-RJ. "
    "Use o contexto fornecido para responder de forma precisa e concisa. "
    "Se voce nao souber a resposta, diga que nao sabe. "
    "Contexto: {context}"
)

prompt = ChatPromptTemplate.from_messages([
    ("system", system_prompt),
    ("human", "{input}"),
])

question_answer_chain = create_stuff_documents_chain(llm, prompt)
qa_chain = create_retrieval_chain(
    db.as_retriever(search_kwargs={"k": 3}),
    question_answer_chain
)

def process_query(query):
    try:
        result = qa_chain.invoke({"input": query})
        answer = result.get("answer", "Nao foi possivel encontrar uma resposta.")
        return answer
    except Exception as e:
        return f"Erro ao processar: {str(e)}"

with gr.Blocks(theme=gr.themes.Soft()) as demo:
    gr.Markdown("# Assistente de Analise do Contrato DETRAN-RJ")
    gr.Markdown("Faca uma pergunta sobre os documentos do contrato.")
    with gr.Row():
        inp = gr.Textbox(label="Sua pergunta", placeholder="Ex: Qual o valor total do contrato?", scale=4)
        btn = gr.Button("Enviar", variant="primary", scale=1)
    out = gr.Markdown(label="Resposta")
    btn.click(fn=process_query, inputs=inp, outputs=out)
    inp.submit(fn=process_query, inputs=inp, outputs=out)
    gr.Examples(
        examples=["Qual o objeto do contrato?", "Qual o valor total?", "Quem e o gestor?", "Qual o prazo de vigencia?"],
        inputs=inp
    )

demo.launch()