""" app.py — Streamlit interface for the Digital KP 2030 assistant. Run locally: streamlit run app.py On Hugging Face Spaces this file is detected and launched automatically. """ from __future__ import annotations import streamlit as st from rag import Assistant, EXAMPLE_QUESTIONS, LLM_MODEL st.set_page_config( page_title="Digital KP 2030 Assistant", page_icon="📘", layout="centered", initial_sidebar_state="expanded", ) # -------------------------------------------------------------------------- # Styling — a restrained navy palette matching the source document # -------------------------------------------------------------------------- st.markdown(""" """, unsafe_allow_html=True) st.markdown("""

Digital KP 2030 Assistant

Answers drawn only from the Khyber Pakhtunkhwa Digital Transformation Policy & Roadmap 2030, with page citations.

""", unsafe_allow_html=True) # -------------------------------------------------------------------------- # Load the index once per session # -------------------------------------------------------------------------- @st.cache_resource(show_spinner="Loading the document index…") def load_assistant() -> Assistant: return Assistant() try: assistant = load_assistant() except FileNotFoundError as exc: st.error(str(exc)) st.stop() except Exception as exc: # noqa: BLE001 st.error(f"Could not start the assistant: {type(exc).__name__}: {exc}") st.stop() # -------------------------------------------------------------------------- # Sidebar # -------------------------------------------------------------------------- with st.sidebar: st.subheader("About") st.write( "This assistant retrieves passages from the published Digital KP 2030 " "document and answers strictly from them. If something is not in the " "document, it says so rather than guessing." ) meta = assistant.meta st.markdown( f'{meta["pages"]} pages' f'{meta["chunks"]} chunks' f'{LLM_MODEL.split("/")[-1]}', unsafe_allow_html=True, ) st.divider() st.subheader("Try a question") for q in EXAMPLE_QUESTIONS[:6]: if st.button(q, key=f"ex-{q}", use_container_width=True): st.session_state.pending = q st.divider() if st.button("Clear conversation", use_container_width=True): st.session_state.messages = [] st.rerun() st.caption( "Source: Khyber Pakhtunkhwa Digital Transformation Board. " "Verify anything consequential against the published PDF." ) # -------------------------------------------------------------------------- # Chat state # -------------------------------------------------------------------------- if "messages" not in st.session_state: st.session_state.messages = [] for msg in st.session_state.messages: with st.chat_message(msg["role"]): st.markdown(msg["content"]) if msg.get("sources"): with st.expander(f"Sources — {len(msg['sources'])} passage(s)"): for s in msg["sources"]: preview = s["text"][:420] + ("…" if len(s["text"]) > 420 else "") st.markdown( f'
Page {s["page"]} ' f'· relevance {s["score"]:.2f}
{preview}
', unsafe_allow_html=True, ) typed = st.chat_input("Ask about goals, pillars, KPIs, governance or timelines…") question = typed or st.session_state.pop("pending", None) if question: st.session_state.messages.append({"role": "user", "content": question}) with st.chat_message("user"): st.markdown(question) with st.chat_message("assistant"): with st.spinner("Searching the document…"): history = [ {"role": m["role"], "content": m["content"]} for m in st.session_state.messages[:-1] ] result = assistant.answer(question, history=history) st.markdown(result["answer"]) if result["error"] and result["error"] != "missing_key": st.warning(f"Model error: {result['error']}") sources = [ {"page": s.page, "text": s.text, "score": s.score} for s in result["sources"] ] if sources: with st.expander(f"Sources — {len(sources)} passage(s)"): for s in sources: preview = s["text"][:420] + ("…" if len(s["text"]) > 420 else "") st.markdown( f'
Page {s["page"]} ' f'· relevance {s["score"]:.2f}
{preview}
', unsafe_allow_html=True, ) st.session_state.messages.append({ "role": "assistant", "content": result["answer"], "sources": sources, })