app.py

# ✅ Arabic RAG System with Cascaded Memory Pipeline
# Step-by-step: Chunking -> Question -> Retrieval -> Summarization

import gradio as gr
from rag_pipeline import RAGPipeline
from utils import process_documents

rag = RAGPipeline()

def log_message(msg, logs):
    logs = logs + msg + "\n"
    return logs

def upload_and_prepare(files, logs):
    logs = log_message("[RAG] بدء معالجة الملفات...", logs)
    all_chunks = []
    for file in files:
        logs = log_message(f"[RAG] معالجة الملف: {file.name}", logs)
        chunks = process_documents(file.name)
        all_chunks.extend(chunks)
        logs = log_message(f"[RAG] تم استخراج {len(chunks)} مقطع من {file.name}", logs)

    rag.build_index(all_chunks)
    logs = log_message(f"[RAG] تم بناء الفهرس بـ {len(all_chunks)} مقطع.", logs)
    return all_chunks, logs, gr.update(visible=True), gr.update(visible=True), gr.update(visible=True)

def store_question(q):
    return q

def retrieve_passages(question, chunks, logs):
    passages = rag.retrieve_passages(question)
    passages_text = "\n---\n".join(passages)
    logs = log_message(f"[RAG] تم العثور على {len(passages)} مقطع مرتبط بالسؤال.", logs)
    return passages, passages_text, logs

def generate_summary_only(passages_text, logs):
    summary = rag.summarize_text(passages_text)
    if not summary.strip():
        logs = log_message("[RAG] ⚠️ لم يتم توليد ملخص. تحقق من محتوى المقاطع.", logs)
    else:
        logs = log_message("[RAG] تم توليد الملخص بنجاح.\n📌 الملخص:\n" + summary, logs)
    return summary, logs

def generate_final_answer(question, summary_text, logs):
    answer, _ = rag.generate_answer_from_passages(question, summary_text)
    if not answer.strip():
        logs = log_message("[RAG] ⚠️ لم يتم توليد إجابة. ربما النص طويل أو النموذج فشل.", logs)
    else:
        logs = log_message("[RAG] ✅ تم توليد الإجابة النهائية.", logs)
    return answer, logs

with gr.Blocks() as demo:
    logs = gr.State("")
    chunks = gr.State([])
    question_state = gr.State("")
    retrieved_passages = gr.State([])

    gr.Markdown("# 🕌 نظام استرجاع المعرفة - مراحل متسلسلة")

    with gr.Row():
        files_input = gr.File(file_types=[".pdf", ".docx", ".txt"], file_count="multiple")
        upload_btn = gr.Button("⬆️ رفع وتجهيز المقاطع")

    with gr.Row():
        question_input = gr.Textbox(label="❓ اكتب سؤالك")
        store_question_btn = gr.Button("📥 تخزين السؤال")

    find_btn = gr.Button("🔍 بحث عن المقاطع المرتبطة")
    summarize_btn = gr.Button("📝 توليد الملخص")
    answer_btn = gr.Button("✍️ توليد الإجابة النهائية")

    passage_output = gr.Textbox(label="📄 المقاطع المرتبطة", lines=8)
    summary_output = gr.Textbox(label="📌 الملخص المستخدم", lines=5)
    answer_output = gr.Textbox(label="✅ الإجابة النهائية", lines=5)
    logs_output = gr.Textbox(label="📜 سجل العمليات", lines=10, interactive=False)

    upload_btn.click(upload_and_prepare, inputs=[files_input, logs], outputs=[chunks, logs_output, store_question_btn, find_btn, summarize_btn])
    store_question_btn.click(store_question, inputs=question_input, outputs=question_state)
    find_btn.click(retrieve_passages, inputs=[question_state, chunks, logs], outputs=[retrieved_passages, passage_output, logs_output])
    summarize_btn.click(generate_summary_only, inputs=[passage_output, logs], outputs=[summary_output, logs_output])
    answer_btn.click(generate_final_answer, inputs=[question_state, summary_output, logs], outputs=[answer_output, logs_output])

demo.launch()