app.py

import gradio as gr
import pandas as pd
from sentence_transformers import SentenceTransformer
from sklearn.metrics.pairwise import cosine_similarity
from transformers import pipeline

# Charger la FAQ
df = pd.read_csv("faq.csv", names=["question", "reponse"])

# Embeddings pour la similarité
model_embed = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
faq_embeddings = model_embed.encode(df["question"].tolist())

# Générateur (ex. flan-t5)
generator = pipeline("text2text-generation", model="google/flan-t5-base")

def generate_prompt(user_question):
    user_emb = model_embed.encode([user_question])
    similarities = cosine_similarity(user_emb, faq_embeddings)[0]
    top_indices = similarities.argsort()[-3:][::-1]

    context = ""
    for i in top_indices:
        context += f"Q: {df.iloc[i]['question']}\nA: {df.iloc[i]['reponse']}\n\n"

    prompt = f"""Tu es un assistant pour Farasha Systems.

Voici des exemples issus de la FAQ :
{context}
Question : {user_question}
Réponse :"""
    return prompt

# Fonction commune pour les deux interfaces
def chat_fn(user_message, history):
    prompt = generate_prompt(user_message)
    response = generator(prompt, max_new_tokens=256)
    answer = response[0]['generated_text'].strip()
    return answer


# ✅ Interface conversationnelle (type ChatGPT)
chat_ui = gr.ChatInterface(
    fn=chat_fn,
    title="🤖 Farasha Assistant",
    description="Posez une question sur Farasha Systems.",
    theme="soft"
)

# ✅ Interface API REST pour JS
api_interface = gr.Interface(
    fn=chat_fn,
    inputs=gr.Textbox(lines=2, placeholder="Posez votre question ici..."),
    outputs="text",
    api_name="chat"
)

# ✅ Lancement des deux
demo = gr.TabbedInterface([chat_ui, api_interface], tab_names=["💬 Chatbot", "🔧 API"])
demo.launch()