app.py

"""
Streamlit Chat Interface — the primary demo surface.

The challenge spec calls for an "Autonomous Financial Advisor Chat Agent"
that answers any finance/portfolio question. This file provides that UI:

  - Sidebar: portfolio dropdown + compact context (P&L, confidence, drivers)
             + collapsible "Full analysis" expander for audit.
  - Main:    chat window (history + input). Every user message is answered
             using the Briefing as ground truth via src.chat.chat().

The heavy-lift reasoning (Phases 1-4) runs ONCE when a portfolio is
selected. Its output (a `Briefing`) is cached in session state and
reused as context for every chat turn — so follow-up questions are fast
and consistent.

Run:
    streamlit run app.py
"""
from __future__ import annotations

import os
from pathlib import Path

# Load env before importing anything that reads env vars
from dotenv import load_dotenv
load_dotenv()

import streamlit as st

from main import analyse_portfolio
from src.chat import chat
from src.data_loader import DataLoader
from src.models import Briefing
from src.phase4_observability import trace_run

DATA_DIR = Path(__file__).parent / "data" / "json"


# -----------------------------------------------------------------------------
# Data loading (cached)
# -----------------------------------------------------------------------------

@st.cache_resource
def get_loader() -> DataLoader:
    return DataLoader(DATA_DIR)


@st.cache_data(show_spinner=False)
def list_portfolios() -> dict[str, str]:
    loader = get_loader()
    return {
        pid: f"{pid} — {p.user_name} ({p.portfolio_type})"
        for pid, p in loader.load_all_portfolios().items()
    }


# -----------------------------------------------------------------------------
# Page setup
# -----------------------------------------------------------------------------

st.set_page_config(
    page_title="Financial Advisor Chat",
    layout="wide",
    initial_sidebar_state="expanded",
)


# -----------------------------------------------------------------------------
# Sidebar — portfolio picker + compact status + runtime info
# -----------------------------------------------------------------------------

st.sidebar.title("Financial Advisor")
st.sidebar.caption("Causal reasoning over Indian equity portfolios.")

portfolios = list_portfolios()
selected = st.sidebar.selectbox(
    "Portfolio",
    options=list(portfolios.keys()),
    format_func=lambda pid: portfolios[pid],
    index=1,  # default to PORTFOLIO_002 (banking-heavy = richest demo)
)


# -----------------------------------------------------------------------------
# Run analysis when portfolio changes (session-scoped — re-runs only on change)
# -----------------------------------------------------------------------------

if st.session_state.get("current_portfolio") != selected:
    with st.spinner(f"Analysing {selected}..."):
        st.session_state.briefing = analyse_portfolio(get_loader(), selected)
    st.session_state.current_portfolio = selected

    b = st.session_state.briefing
    opening = (
        f"I've analysed **{b.portfolio.user_name}**'s portfolio "
        f"for {b.market.context.date}.\n\n"
        f"**Day P&L:** {b.analytics.total_pnl_percent:+.2f}% "
        f"({b.analytics.total_pnl_inr:+,.0f} INR)  \n"
        f"**Primary driver:** {b.causal.primary_driver}  \n"
        f"**Confidence:** {b.confidence.overall:.2f}\n\n"
        "Ask me anything — I have the full causal analysis in memory. For example:\n\n"
        "- *Why did my portfolio move today?*\n"
        "- *Show me any conflicts in today's signals*\n"
        "- *What if I hadn't held banking?*\n"
        "- *How risky is my portfolio?*\n"
        "- *Tell me about HDFCBANK* (or any holding)\n"
        "- *What is CAPM?* _(general finance — needs Gemini key)_"
    )
    st.session_state.messages = [{"role": "assistant", "content": opening}]


briefing: Briefing = st.session_state.briefing


# -----------------------------------------------------------------------------
# Sidebar — compact status
# -----------------------------------------------------------------------------

st.sidebar.divider()

pnl_col_a, pnl_col_b = st.sidebar.columns(2)
pnl_col_a.metric(
    "Day P&L",
    f"{briefing.analytics.total_pnl_percent:+.2f}%",
    delta=f"{briefing.analytics.total_pnl_inr:+,.0f} INR",
)
pnl_col_b.metric("Confidence", f"{briefing.confidence.overall:.2f}")

st.sidebar.markdown(f"**Primary driver:** {briefing.causal.primary_driver}")
if briefing.market.active_macro_themes:
    themes = briefing.market.active_macro_themes
    shown = ", ".join(themes[:3])
    if len(themes) > 3:
        shown += f" _(+{len(themes)-3} more)_"
    st.sidebar.caption(f"**Themes:** {shown}")

if briefing.analytics.concentration_risks:
    st.sidebar.warning(
        briefing.analytics.concentration_risks[0],
        icon=":material/warning:",
    )


# -----------------------------------------------------------------------------
# Sidebar — full analysis (collapsed by default)
# -----------------------------------------------------------------------------

with st.sidebar.expander("Full analysis", expanded=False):
    st.markdown("**Causal chain** _(themes touching your portfolio)_")
    impacts_by_sector: dict[str, list] = {}
    for i in briefing.causal.stock_impacts:
        impacts_by_sector.setdefault(i.sector, []).append(i)
    for theme, sectors in briefing.causal.macro_to_sector.items():
        hit = [s for s in sectors if s in impacts_by_sector]
        if not hit:
            continue
        st.markdown(f"- **{theme}** → {', '.join(hit[:4])}")

    if briefing.causal.stock_impacts:
        st.markdown("**Top drivers**")
        for i in briefing.causal.stock_impacts[:3]:
            st.caption(
                f"{i.ticker} · {i.day_change_pct:+.2f}% · "
                f"contrib {i.pnl_contribution_pct:+.2f}pp · β={i.beta}"
            )

    if briefing.causal.counterfactuals:
        st.markdown("**Counterfactuals**")
        for cf in briefing.causal.counterfactuals:
            st.caption(
                f"{cf.scenario}: {cf.hypothetical_pnl_pct:+.2f}% "
                f"(delta {cf.delta_pct:+.2f}pp)"
            )

    st.markdown("**Self-evaluation**")
    c = briefing.confidence
    st.caption(
        f"data {c.data_completeness:.2f} · news {c.news_coverage:.2f} · "
        f"signal {c.signal_strength:.2f} · conflict {c.conflict_penalty:.2f} · "
        f"reasoning {c.reasoning_quality:.2f}"
    )


# -----------------------------------------------------------------------------
# Sidebar — runtime indicators
# -----------------------------------------------------------------------------

st.sidebar.divider()
llm_on = os.getenv("USE_LLM", "true").lower() == "true" and bool(
    os.getenv("GROQ_API_KEY")
)
langfuse_on = bool(os.getenv("LANGFUSE_PUBLIC_KEY") and os.getenv("LANGFUSE_SECRET_KEY"))

st.sidebar.markdown("**Runtime**")
st.sidebar.caption(
    f"LLM: {'openai/gpt-oss-120b' if llm_on else 'keyword fallback (portfolio questions only)'}"
)
st.sidebar.caption(
    
    f"Tracing: {'Langfuse + stderr JSON' if langfuse_on else 'stderr JSON only'}"
)
if not llm_on:
    st.sidebar.caption(
        "_Set `GROQ_API_KEY` + `USE_LLM=true` in `.env` to enable general finance Q&A._"
    )


# -----------------------------------------------------------------------------
# Main — chat window
# -----------------------------------------------------------------------------

st.title("Financial Advisor Chat")
st.caption(
    f"Context: **{briefing.portfolio.user_name}** · "
    f"{briefing.portfolio.portfolio_type} · {briefing.market.context.date}"
)

# Render chat history
for msg in st.session_state.messages:
    with st.chat_message(msg["role"]):
        st.markdown(msg["content"])
        # Show attribution footer on assistant messages when present
        if msg["role"] == "assistant" and msg.get("meta"):
            st.caption(msg["meta"])


# -----------------------------------------------------------------------------
# Chat input
# -----------------------------------------------------------------------------

if prompt := st.chat_input("Ask about your portfolio or any finance question..."):
    st.session_state.messages.append({"role": "user", "content": prompt})
    with st.chat_message("user"):
        st.markdown(prompt)

    with st.chat_message("assistant"):
        with st.spinner("Thinking..."):
            with trace_run(
                name=f"chat_{selected}",
                metadata={"portfolio_id": selected, "query": prompt[:120]},
            ):
                response, used_llm, tin, tout, lat = chat(
                    prompt,
                    briefing,
                    history=st.session_state.messages[:-1],
                )
        st.markdown(response)
        meta_caption = (
            f"{tin}→{tout} tokens · {lat:.0f} ms · via Groq (openai/gpt-oss-120b)"
            if used_llm
            else "via keyword fallback · enable Gemini for broader answers"
        )
        st.caption(meta_caption)

    st.session_state.messages.append({
        "role": "assistant",
        "content": response,
        "meta": meta_caption,
    })