app.py

import gradio as gr
import spaces
import faiss
import numpy as np
from datasets import load_dataset
from sentence_transformers import SentenceTransformer
from transformers import RagTokenizer, RagSequenceForGeneration

# — Config —
DATASET_NAME = "username/mealplan-chunks"
INDEX_PATH   = "mealplan.index"
MODEL_NAME   = "facebook/rag-sequence-nq"

# — Load chunks & FAISS index —
ds = load_dataset(DATASET_NAME, split="train")
texts   = ds["text"]
sources = ds["source"]
pages   = ds["page"]

# — Embeddings embedder & FAISS —
embedder         = SentenceTransformer("all-MiniLM-L6-v2")
chunk_embeddings = embedder.encode(texts, convert_to_numpy=True)
index            = faiss.read_index(INDEX_PATH)

# — RAG generator —
tokenizer = RagTokenizer.from_pretrained(MODEL_NAME)
model     = RagSequenceForGeneration.from_pretrained(MODEL_NAME)

@spaces.GPU
def respond(
    message: str,
    history: list[tuple[str, str]],
    goal: str,
    diet: list[str],
    meals: int,
    avoid: str,
    weeks: str,
):
    # Parse preferences
    avoid_list = [a.strip() for a in avoid.split(",") if a.strip()]
    prefs = (
        f"Goal={goal}; Diet={','.join(diet)}; "
        f"Meals={meals}/day; Avoid={','.join(avoid_list)}; Duration={weeks}"
    )

    # 1) Query embedding & FAISS search
    q_emb = embedder.encode([message], convert_to_numpy=True)
    D, I  = index.search(q_emb, 5)  # top-5
    ctx_chunks = [
        f"[{sources[i]} p{pages[i]}] {texts[i]}" for i in I[0]
    ]
    context = "\n".join(ctx_chunks)

    # 2) Build prompt
    prompt = (
        "SYSTEM: Only answer using the provided CONTEXT. "
        "If it’s not there, say \"I'm sorry, I don't know.\"\n"
        f"PREFS: {prefs}\n"
        f"CONTEXT:\n{context}\n"
        f"Q: {message}\n"
    )

    # 3) Generate
    inputs  = tokenizer([prompt], return_tensors="pt")
    outputs = model.generate(**inputs, num_beams=2, max_new_tokens=200)
    answer  = tokenizer.batch_decode(outputs, skip_special_tokens=True)[0]

    # 4) Update history
    history = history or []
    history.append((message, answer))
    return history

# — Build Gradio chat interface —
goal  = gr.Dropdown(["Lose weight","Bulk","Maintain"], value="Lose weight", label="Goal")
diet  = gr.CheckboxGroup(["Omnivore","Vegetarian","Vegan","Keto","Paleo","Low-Carb"], label="Diet Style")
meals = gr.Slider(1,6,value=3,step=1,label="Meals per day")
avoid = gr.Textbox(placeholder="e.g. Gluten, Dairy, Nuts...", label="Avoidances (comma-separated)")
weeks = gr.Dropdown(["1 week","2 weeks","3 weeks","4 weeks"], value="1 week", label="Plan Length")

demo = gr.ChatInterface(
    fn=respond,
    additional_inputs=[goal, diet, meals, avoid, weeks],
)

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