BI Platform

End-to-End Business Intelligence — Streamlit + Plotly + SQLite

Six analytical dashboards. Zero cloud dependencies. Fully interactive.

---

What It Does

A fully self-contained BI platform that demonstrates end-to-end dashboard engineering:

Dashboard	Purpose
🏠 Overview	KPI cards, monthly trend + MoM growth, region donut, category bars
📈 Revenue	Monthly/quarterly trends, region area chart, tabbed drill-in
📦 Products	Category bars, return-rate scatter, cohort category heatmap
👤 Sales Team	Leaderboard (colour-coded by quota status), deal-size scatter
🔍 Drill-Down	Region/category/month filters → raw order table + CSV export
🔄 Returns	Return-rate heatmap by category × region

---

Quick Start

git clone https://github.com/YOUR_USERNAME/bi-platform.git
cd bi-platform
python -m venv venv && venv\Scripts\activate   # Windows
pip install -r requirements.txt

# Run tests (no API key / no cloud credentials needed)
pytest tests/ -v

# Seed the database and launch
python -m src.database            # seeds data/bi_platform.db
streamlit run src/app.py
# Open http://localhost:8501

---

Architecture

flowchart LR
    A["SQLite DB\n2,000 synthetic rows\n(auto-seeded)"] --> B["analytics.py\nQuery layer\n(pure pandas)"]
    B --> C["charts.py\nPlotly builders\n(pure figures)"]
    C --> D["app.py\nStreamlit multi-page\n(6 pages / tabs)"]
    D --> E["Browser\nInteractive dashboard"]

Loading

Three-layer design:

Layer	File	Role
Data	src/database.py	SQLite schema + deterministic seed (2 k rows, seed=42)
Analytics	src/analytics.py	SQL → pandas, KPIs, cohorts, returns
Presentation	src/charts.py + src/app.py	Plotly figures + Streamlit pages

---

Database Schema

products     (product_id PK, name, category, unit_price)
salespeople  (salesperson_id PK, name, region)
sales        (sale_id PK, sale_date, product_id FK, salesperson_id FK,
              region, quantity, unit_price, revenue, is_returned)

• 2,000 rows generated deterministically (seed=42) — identical data on every machine
• Q4 seasonal uplift baked in (40 % higher quantity multiplier)
• 4 % return rate injected randomly

---

Key Techniques

• Watermark-free SQLite — no ORM, raw sqlite3 + pandas read_sql_query for fast iteration
• Three-layer separation — DB → analytics → charts layers are independently testable
• Streamlit @st.cache_data — query results cached 5 min so navigation doesn't re-query
• Plotly dual-axis — revenue bars + MoM growth % line share one figure (secondary y-axis)
• Pivot cohort heatmap — df.pivot + row-wise normalisation → percentage share per month

---

What I Learned

• Multi-page Streamlit — sidebar st.radio as a router, each page function is independently callable
• Plotly dual y-axis — yaxis2 with overlaying="y" and side="right" in update_layout
• SQLite for BI prototyping — CASE expressions, sub-selects, and window-like patterns without a heavyweight DB
• Test isolation — tmp_path_factory session-scoped fixture gives one seeded DB for all tests, no I/O per test
• Dependency inversion — db_path parameter on every analytics function enables test DBs without monkeypatching

---

License

MIT

_{Part of a 10-project Data Analyst portfolio}