app.py

# -*- coding: utf-8 -*-
"""app

Automatically generated by Colab.

Original file is located at
    https://colab.research.google.com/drive/1PhcQoTZvxdPQe6E1HMx_Nl4Zs_tY7J_y
"""



import gradio as gr
import os
from dotenv import load_dotenv
from langchain.chat_models import ChatOpenAI
from langchain.document_loaders import PyPDFLoader
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.vectorstores import FAISS
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.chains import RetrievalQA
import tempfile
import numpy as np
from sklearn.metrics.pairwise import cosine_similarity

# ✅ Carrega variáveis de ambiente
load_dotenv()
OPENROUTER_API_KEY = os.getenv("ROUTER_API_KEY")

if not OPENROUTER_API_KEY:
    raise ValueError("❌ A variável de ambiente ROUTER_API_KEY não está definida. Verifique o arquivo .env.")

# Inicialização
embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
qa_chain = None
processed_file = None

# 🔁 Carrega automaticamente o legisMiner.pdf ao iniciar
def load_default_pdf():
    global qa_chain, processed_file
    try:
        loader = PyPDFLoader("LegisMiner.pdf")
        documents = loader.load()

        text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
        docs = text_splitter.split_documents(documents)

        vectorstore = FAISS.from_documents(docs, embeddings)

        llm = ChatOpenAI(
            openai_api_key=OPENROUTER_API_KEY,
            openai_api_base="https://openrouter.ai/api/v1",
            model="deepseek/deepseek-r1-0528:free",
            temperature=0.7
        )

        qa_chain = RetrievalQA.from_chain_type(
            llm=llm,
            retriever=vectorstore.as_retriever(),
            return_source_documents=True
        )

        processed_file = "LegisMiner.pdf"
        print("✅ LegisMiner.pdf carregado automaticamente.")
    except Exception as e:
        print(f"❌ Erro ao carregar LegisMiner.pdf automaticamente: {e}")

def calculate_rag_metrics(query, response, source_docs):
    metrics = {}
    try:
        query_embedding = embeddings.embed_query(query)
        response_embedding = embeddings.embed_query(response)
        metrics["query_response_similarity"] = cosine_similarity(
            [query_embedding], [response_embedding]
        )[0][0]

        doc_similarities = []
        for doc in source_docs:
            doc_embedding = embeddings.embed_query(doc.page_content[:1000])
            similarity = cosine_similarity([response_embedding], [doc_embedding])[0][0]
            doc_similarities.append(similarity)

        metrics["avg_response_source_similarity"] = np.mean(doc_similarities) if doc_similarities else 0
        metrics["max_response_source_similarity"] = max(doc_similarities) if doc_similarities else 0
        metrics["num_source_documents"] = len(source_docs)

    except Exception as e:
        metrics["error"] = str(e)

    return metrics

def process_pdf(file):
    global qa_chain, processed_file

    with tempfile.NamedTemporaryFile(delete=False, suffix=".pdf") as tmp:
        tmp.write(file)
        pdf_path = tmp.name

    try:
        loader = PyPDFLoader(pdf_path)
        documents = loader.load()

        text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
        docs = text_splitter.split_documents(documents)

        vectorstore = FAISS.from_documents(docs, embeddings)

        llm = ChatOpenAI(
            openai_api_key=OPENROUTER_API_KEY,
            openai_api_base="https://openrouter.ai/api/v1",
            model="deepseek/deepseek-r1-0528-qwen3-8b:free",
            temperature=0.7
        )

        qa_chain = RetrievalQA.from_chain_type(
            llm=llm,
            retriever=vectorstore.as_retriever(),
            return_source_documents=True
        )

        processed_file = os.path.basename(pdf_path)
        return f"✅ PDF processado com sucesso: {processed_file}"

    except Exception as e:
        return f"❌ Erro ao processar PDF: {str(e)}"

def ask_question(question):
    global qa_chain

    if qa_chain is None:
        return "⚠️ Por favor, carregue um PDF primeiro", "", {}

    try:
        # ✅ Prompt especializado: Analista de Normas Regulatórias de Mineração
        system_prompt = (
            "Você é um Analista Especialista em Normas Regulatórias de Mineração no Brasil. "
            "Baseie suas respostas exclusivamente no conteúdo técnico do documento LegisMiner.pdf. "
            "Forneça respostas claras, técnicas e fundamentadas. "
            "Se a informação não estiver presente no documento, informe isso ao usuário."
        )

        resposta = qa_chain.invoke({
            "query": f"{system_prompt}\n\nPergunta: {question}"
        })

        sources = "\n\n".join(
            [f"📄 Fonte {i+1}:\n{doc.page_content[:500]}..."
             for i, doc in enumerate(resposta['source_documents'])]
        )

        metrics = calculate_rag_metrics(
            question,
            resposta['result'],
            resposta['source_documents']
        )

        metrics_text = "\n".join(
            [f"{k.replace('_', ' ').title()}: {v:.2f}" if isinstance(v, float) else f"{k.replace('_', ' ').title()}: {v}"
             for k, v in metrics.items() if k != "error"]
        )

        return resposta['result'], sources, metrics_text

    except Exception as e:
        return f"❌ Erro ao processar pergunta: {str(e)}", "", {}

# Interface Gradio
with gr.Blocks(title="Chat com PDF usando OpenRouter", theme=gr.themes.Soft()) as demo:
    gr.Markdown("# 🧠 Artificial Intelligence Applied to Regulatory Standard Processing in Mining\n### 💡 Development of a Decision Support Tool")

    with gr.Row():
        with gr.Column(scale=1):
            file_input = gr.File(label="📤 Envie um PDF", type="binary")
            process_btn = gr.Button("Processar PDF", variant="primary")
            status_output = gr.Textbox(label="Status")

        with gr.Column(scale=2):
            question_input = gr.Textbox(label="Faça uma pergunta sobre Normas da Mineração", lines=3)
            ask_btn = gr.Button("Enviar Pergunta", variant="primary")
            answer_output = gr.Textbox(label="✅ Resposta", interactive=False)

        with gr.Accordion("📄 Fontes usadas", open=False):
            sources_output = gr.Textbox(label="Trechos relevantes", lines=10)

        with gr.Accordion("📊 Métricas RAG", open=False):
            metrics_output = gr.Textbox(label="Métricas", lines=4)

    process_btn.click(
        fn=process_pdf,
        inputs=file_input,
        outputs=status_output
    )

    ask_btn.click(
        fn=ask_question,
        inputs=question_input,
        outputs=[answer_output, sources_output, metrics_output]
    )

# 🔁 Carrega o PDF fixo ao iniciar
load_default_pdf()

# Compartilhamento opcional no Colab ou Hugging Face
share = True if 'COLAB_JUPYTER_TRANSPORT' in os.environ else False
demo.launch(share=share, debug=False)