app_back.py

import spaces

import gradio as gr

from src.vectorstore.faiss_index import load_index
from src.vectorstore.metadata_store import load_metadata

from src.retrieval.retriever import retrieve
from src.retrieval.reranker import rerank

from src.llm.generator import generate_answer
from src.llm.postprocessing import clean_answer

from src.data_processing.txt_embeddings import compute_embeddings

from src.voice.assistant import text_to_audio

# Load once
from pathlib import Path

BASE_DIR = Path(__file__).parent

INDEX_PATH = str(BASE_DIR / "data/vectorstore/faiss.index")
METADATA_PATH = str(BASE_DIR / "data/metadata/metadata.json")

index = load_index(INDEX_PATH)
metadata = load_metadata(METADATA_PATH)

@spaces.GPU


def chat_fn(message, history):

    # 1. Embedding de la question
    query_emb = compute_embeddings([message])

    # 2. Retrieval FAISS
    docs = retrieve(query_emb, index, metadata, top_k=10)

    # 3. Reranking (IMPORTANT: retourne (doc, score))
    docs = rerank(message, docs, top_k=10)

    # 4. Construction du contexte LLM
    context = "\n\n".join([doc["text"] for doc, score in docs])

    # 5. Prompt
    prompt = f"""
Context:
{context}

Question:
{message}

Answer:
"""

    # 6. Génération réponse LLM
    answer = generate_answer(prompt)

    # 6.1. Postprocessing
    answer = clean_answer(answer)

    # 6.2. Audio answer
    audio = text_to_audio(answer)

    # 7. Debug chunks avec scores
    debug = "\n\n".join(
        [
            f"📄 Chunk {i+1} (Score : {score:.3f})\n{doc['text']}"
            for i, (doc, score) in enumerate(docs)
        ]
    )

    # 8. Output final
    final_output = f"{answer}\n\n---\n\n🔍 Retrieved chunks:\n{debug}"


    return final_output, audio

###########################################################################
# Interface Gradio type ChatGPT

"""

audio_output = gr.Audio(
    label="🔊 Audio"
)

demo = gr.ChatInterface(
    fn=chat_fn,
    title="💬 RAG Assistant",
    description="Chat avec ton système RAG local",
    additional_outputs=[audio_output]
)

if __name__ == "__main__":
    demo.launch()
"""