app.py

import gradio as gr
import pandas as pd
import faiss
from sentence_transformers import SentenceTransformer
from transformers import pipeline

# Load abstracts
df = pd.read_csv("abstracts.csv")
abstracts = df["abstract"].tolist()

# Load embedding model
embedder = SentenceTransformer("all-MiniLM-L6-v2")
abstract_embeddings = embedder.encode(abstracts, show_progress_bar=True)

# Build FAISS index
index = faiss.IndexFlatL2(abstract_embeddings.shape[1])
index.add(abstract_embeddings)

# Load LLM from Hugging Face Hub
llm = pipeline("text-generation", model="tiiuae/falcon-7b-instruct", max_new_tokens=300)

def verify_claim(claim):
    query_vec = embedder.encode([claim])
    D, I = index.search(query_vec, 3)

    top_abstracts = df.iloc[I[0]]["abstract"].tolist()
    context = "\n".join(top_abstracts)

    prompt = f"Claim: {claim}\n\nEvidence:\n{context}\n\nAnswer True, False, or Uncertain. Then explain why:\n"
    output = llm(prompt)[0]["generated_text"]

    return f"🔍 **Top Abstracts:**\n{context}\n\n🧠 **LLM Response:**\n{output}"

# Gradio UI
gr.Interface(
    fn=verify_claim,
    inputs=gr.Textbox(label="Enter a scientific claim"),
    outputs=gr.Markdown(),
    title="🔬 Scientific Claim Verifier",
    description="Checks the validity of a scientific claim using PubMed abstracts + LLM"
).launch()