app.py

"""app.py — BizIntel AI Ultra (Gemini‑only, v4)
Robust BI copilot featuring:
• CSV / Excel / Parquet uploads + live SQL ingestion
• Memory‑safe sampling for large files (≤ 5 M rows sample)
• Gemini‑generated narrative insights & quick schema audit
• Interactive EDA (histogram + box, correlation heatmap)
• Tunable ARIMA forecasting (p,d,q & horizon)
• One‑click strategy brief download
"""
from __future__ import annotations
import io, os, tempfile
from pathlib import Path
from typing import List

import pandas as pd
import plotly.express as px
import streamlit as st
from sqlalchemy import create_engine
from statsmodels.tsa.arima.model import ARIMA
import google.generativeai as genai

# ───────────────────── 0 · PAGE CONFIG ──────────────────────
st.set_page_config(
    page_title="BizIntel AI Ultra",
    page_icon="📊",
    layout="wide",
    initial_sidebar_state="expanded",
)

auth_key = None
try:
    auth_key = st.secrets["GEMINI_APIKEY"]  # type: ignore[attr-defined]
except Exception:
    auth_key = os.getenv("GEMINI_APIKEY")

if not auth_key:
    st.error(
        "❌ **GEMINI_APIKEY** missing. Add it in *Settings → Secrets* or set env var before running."
    )
    st.stop()

genai.configure(api_key=auth_key)
GEM_MODEL = "gemini-1.5-pro-latest"
TMP = Path(tempfile.gettempdir())

# ───────────────────── 1 · HELPERS ──────────────────────────
@st.cache_data(show_spinner=False)
def load_file(buf: io.BufferedReader, sample: bool = False) -> pd.DataFrame:
    suf = Path(buf.name).suffix.lower()
    if suf in {".xls", ".xlsx"}:
        return pd.read_excel(buf, engine="openpyxl")
    if suf == ".parquet":
        return pd.read_parquet(buf)
    return pd.read_csv(buf, nrows=5_000_000 if sample else None)

@st.cache_data(show_spinner=False)
def list_sql_tables(uri: str) -> List[str]:
    return create_engine(uri).table_names()

@st.cache_data(show_spinner=True)
def fetch_sql_table(uri: str, tbl: str) -> pd.DataFrame:
    return pd.read_sql_table(tbl, create_engine(uri))

@st.cache_data(show_spinner=False)
def gemini(prompt: str) -> str:
    return genai.GenerativeModel(GEM_MODEL).generate_content(prompt).text.strip()

# ───────────────────── 2 · DATA LOAD UI ─────────────────────
st.title("📊 BizIntel AI Ultra — Gemini 1.5 Pro BI Copilot")
mode = st.sidebar.radio("Data source", ["Upload file", "SQL DB"], horizontal=True)

df: pd.DataFrame = pd.DataFrame()
if mode == "Upload file":
    upl = st.sidebar.file_uploader("CSV / Excel / Parquet", ["csv", "xls", "xlsx", "parquet"], help="≤ 2 GB")
    sample = st.sidebar.checkbox("Sample first 5 M rows only")
    if upl:
        df = load_file(upl, sample)
else:
    uri = st.sidebar.text_input("SQLAlchemy URI")
    if uri:
        tbl = st.sidebar.selectbox("Table", list_sql_tables(uri))
        if tbl:
            df = fetch_sql_table(uri, tbl)

if df.empty:
    st.info("⬅️ Load data to start analysis.")
    st.stop()

st.success("✅ Data loaded")
st.dataframe(df.head(), use_container_width=True)

# ───────────────────── 3 · OVERVIEW & GEMINI INSIGHT ─────────
rows, cols = df.shape
missing_pct = df.isna().sum().sum() / (rows * cols) * 100
m1,m2,m3 = st.columns(3)
m1.metric("Rows", f"{rows:,}")
m2.metric("Columns", cols)
m3.metric("Missing %", f"{missing_pct:.1f}")

st.subheader("🧠 Gemini Insights")
with st.spinner("Generating narrative…"):
    summ_json = df.describe(include="all", datetime_is_numeric=True).round(2).to_json()
    st.markdown(gemini(
        "You are a senior BI analyst. Provide five bullet insights and three actions for this dataset:\n\n" + summ_json
    ))

# ───────────────────── 4 · TIME‑SERIES PREP ──────────────────
# attempt datetime coercion
for col in df.columns:
    if not pd.api.types.is_datetime64_any_dtype(df[col]):
        try:
            df[col] = pd.to_datetime(df[col])
        except: pass

DATE_COL = st.selectbox("Date column", [c for c in df.columns if pd.api.types.is_datetime64_any_dtype(df[c])])
NUM_COL = st.selectbox("Numeric metric", [c for c in df.select_dtypes("number").columns])

ts = df[[DATE_COL, NUM_COL]].dropna().groupby(DATE_COL)[NUM_COL].mean().sort_index()
fig_tr = px.line(ts, title=f"{NUM_COL} Trend", labels={"index":"Date", NUM_COL:NUM_COL})
st.plotly_chart(fig_tr, use_container_width=True)

# ───────────────────── 5 · ARIMA FORECAST ───────────────────
st.subheader("🔮 ARIMA Forecast")
steps = st.slider("Horizon (days)", 3, 365, 90)
p = st.number_input("p", 0, 5, 1); d = st.number_input("d", 0, 2, 1); q = st.number_input("q", 0, 5, 1)
try:
    model = ARIMA(ts, order=(p, d, q)).fit()
    future_idx = pd.date_range(ts.index[-1], periods=steps + 1, freq=pd.infer_freq(ts.index) or "D")[1:]
    forecast = pd.Series(model.forecast(steps), index=future_idx, name="Forecast")
    fig_fc = px.line(pd.concat([ts, forecast], axis=1), title="Actual vs Forecast")
    st.plotly_chart(fig_fc, use_container_width=True)
except Exception as e:
    st.error(f"ARIMA failed: {e}")

# ───────────────────── 6 · EDA EXPANDERS ────────────────────
st.subheader("🔍 EDA Dashboard")
with st.expander("Histogram + Box"):
    st.plotly_chart(px.histogram(df, x=NUM_COL, marginal="box", template="plotly_dark"), use_container_width=True)
with st.expander("Correlation Heatmap"):
    corr = df.select_dtypes("number").corr()
    st.plotly_chart(px.imshow(corr, color_continuous_scale="RdBu", title="Correlation"), use_container_width=True)

# ───────────────────── 7 · STRATEGY BRIEF ───────────────────
brief = (
    "# Strategy Brief\n"
    "• Clean missing date values for stable models.\n"
    "• Investigate high correlations for driver analysis.\n"
    "• Use forecast for inventory & workforce planning.\n"
    "• Monitor outliers (>3σ) weekly.\n"
    "• Segment by region/product for targeted actions."
)
st.download_button("⬇️ Download Strategy (.md)", brief, "bizintel_brief.md", "text/markdown")