streamlit_app.py

import streamlit as st
import tempfile

from extractor.pdf_parser import extract_text_from_pdf
from utils.validators import is_text_based_pdf
from chunker.text_chunker import pages_to_chunks
from embeddings.embedder import Embedder
from embeddings.similarity import top_k_similar_chunks
from utils.context_builder import build_context
from llm.explainer import PaperExplainer
from llm.question_bank import QUESTION_BANK

st.set_page_config(page_title="AI Research Paper Explainer", layout="wide")

st.title("📄 AI Research Paper Explainer")
st.caption("Built with Hugging Face · CPU-only · No paid APIs")

uploaded_file = st.file_uploader(
    "Upload a text-based research paper (PDF)",
    type=["pdf"]
)

mode = st.selectbox(
    "Choose analysis mode",
    options=list(QUESTION_BANK.keys())
)

if uploaded_file:
    with tempfile.NamedTemporaryFile(delete=False, suffix=".pdf") as tmp:
        tmp.write(uploaded_file.read())
        pdf_path = tmp.name

    with st.spinner("Extracting and analyzing paper..."):
        pages = extract_text_from_pdf(pdf_path)

        if not is_text_based_pdf(pages):
            st.error("❌ This appears to be a scanned PDF. Please upload a text-based paper.")
            st.stop()

        chunks = pages_to_chunks(pages)

        embedder = Embedder()
        chunk_embeddings = embedder.embed(chunks)

        query_embedding = embedder.embed([pages[0]])

        top_chunks = top_k_similar_chunks(
            query_embedding,
            chunk_embeddings,
            chunks,
            k=5
        )

        context = build_context(top_chunks)

        explainer = PaperExplainer()
        question = QUESTION_BANK[mode]
        answer = explainer.explain(context, question)

    st.subheader(f"🧠 Result — {mode.upper()}")
    st.write(answer)

    with st.expander("🔍 Retrieved Context (Top-K Chunks)"):
        for i, c in enumerate(top_chunks, 1):
            st.markdown(f"**Chunk {i} (score: {c['score']:.2f})**")
            st.write(c["text"][:500])