End-to-end data science and AI automation project covering 7 Melbourne northern corridor suburbs across 12 years (2014–2026).
Built by Khoshaba Odeesho | Assyrian AI
→ View Live on Streamlit Cloud
A fully automated property market intelligence platform that demonstrates:
- SQL data engineering — 6-table PostgreSQL schema with window functions, CAGR calculations, rankings, and permanent views
- Python data science — data cleaning, feature engineering, and ML model training
- Machine learning — XGBoost and Linear Regression price prediction models (XGBoost MAE: $11,744 on real 2024 data)
- AI integration — Google Gemini 2.0 Flash generating professional market narrative reports from real data
- Streamlit dashboard — 5-page interactive dashboard deployed publicly on Streamlit Cloud
- Supabase PostgreSQL — cloud database with 6 tables, 249 rows of real data, live connection
| Suburb | 2014 Price | 2026 Price | CAGR |
|---|---|---|---|
| Mickleham | $168,000 | $710,300 | 12.29% pa |
| Kalkallo | $180,000 | $650,100 | 12.22% pa |
| Beveridge | $351,500 | $642,000 | 6.04% pa |
| Craigieburn | $362,000 | $730,000 | 5.13% pa |
| Wallan | $356,000 | $640,000 | 4.73% pa |
| Roxburgh Park | $385,000 | $630,000 | 4.59% pa |
| Donnybrook | $530,000 | $650,000 | 3.46% pa |
Victorian Gov Data (2014–2024)
Supplementary Sources (2025–2026)
│
▼
Supabase PostgreSQL
┌─────────────────────────────────────┐
│ prop_historical_prices (91 rows) │
│ prop_suburb_stats (83 rows) │
│ prop_ml_predictions (14 rows) │
│ prop_ai_reports (8 rows) │
│ prop_rent_snapshot (7 rows) │
│ prop_live_listings (46 rows) │
│ prop_cagr_summary [VIEW] │
└─────────────────────────────────────┘
│
┌────┴────┐
▼ ▼
Python ML Gemini AI
XGBoost Narratives
LinearReg 8 Reports
│ │
└────┬────┘
▼
Streamlit Dashboard
5 Pages · Live · Public
| Page | What It Shows |
|---|---|
| 🏠 Overview | Corridor KPIs, price history, CAGR leaderboard |
| 📈 Price Intelligence | 12-year trends, YoY heatmap, suburb deep-dive |
| 🤖 ML Predictions | XGBoost vs Linear Regression comparison |
| 🏘️ Live Listings | Current properties for sale and rent |
| 📋 AI Reports | Gemini-generated market narratives |
| Model | MAE | R² | Verdict |
|---|---|---|---|
| XGBoost v1 | $11,744 | 0.3189 | ✅ Champion |
| Linear Regression v1 | $41,584 | -5.27 | Baseline |
Best prediction: Donnybrook — only 0.42% error
Average XGBoost confidence: 98.1%
| Layer | Tool |
|---|---|
| Database | Supabase PostgreSQL |
| Data Science | Python, pandas, numpy |
| ML Models | scikit-learn, XGBoost |
| AI Narratives | Google Gemini 2.0 Flash API |
| Dashboard | Streamlit + Plotly |
| Secrets | python-dotenv |
| Version Control | Git + GitHub |
prop-market-intelligence/
├── prop_dashboard.py # Streamlit 5-page dashboard
├── prop_ml_model.py # XGBoost + Linear Regression training
├── prop_ai_narrative.py # Gemini AI report generation
├── requirements.txt # Python dependencies
├── .gitignore # Excludes .env and secrets
└── README.md # This file
# Clone the repo
git clone https://github.com/Assyrian91/prop-market-intelligence.git
cd prop-market-intelligence
# Install dependencies
pip install -r requirements.txt
# Create .env file with your Supabase credentials
# DB_HOST=your-host
# DB_PORT=5432
# DB_NAME=postgres
# DB_USER=postgres
# DB_PASSWORD=your-password
# Run the dashboard
streamlit run prop_dashboard.py| Source | Coverage | Trust |
|---|---|---|
| Victorian Government Property Sales Report | 2014–2024 all suburbs | ✅ Official |
| SKAD Real Estate | Craigieburn 2025–2026 | ✅ Sourced |
| Valuer-General Victoria | Beveridge 2025 | ✅ Official |
| VicPropertyCheck | Wallan 2026 | ✅ Sourced |
| Picki LGA Data | Mickleham, Kalkallo, Roxburgh Park 2026 | ✅ ABS-sourced |
| Estimated (corridor trend) | 7 sparse rows |
Khoshaba Odeesho
Data Analyst | AI Automation Engineer
Melbourne, Australia
GitHub · LinkedIn
Built as part of a professional data science portfolio — demonstrating end-to-end capability from raw data to deployed AI product.