app.py

import gradio as gr
import pickle
import faiss
from sentence_transformers import SentenceTransformer

# Load SBERT model
model = SentenceTransformer("sentence-transformers/all-mpnet-base-v2")

# Load passages + FAISS index from prebuilt file
with open("sbert_index.pkl", "rb") as f:
    data = pickle.load(f)

passages = data["passages"]
index = data["index"]

def sbert_search(query, k=10):
    query_embedding = model.encode([query])
    distances, indices = index.search(query_embedding, k)
    results = [passages[i] for i in indices[0]]
    return "\n\n".join(results)

# Gradio UI
demo = gr.Interface(
    fn=sbert_search,
    inputs=gr.Textbox(label="Enter your query"),
    outputs=gr.Textbox(label="Top results"),
    title="SBERT Semantic Search",
    description="Search 10,000 MS MARCO passages using SBERT + FAISS"
)

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