AI & Data Engineer — Transport & Urban Mobility · Milan, Italy · abadeanlou.com
I build production ML systems and real-time data pipelines end to end — and I know the transport domain from the inside: four-step demand models, production CUBE migrations, and first-author research on transit accessibility equity.
At CitiEU — CitiME I built and operate a traffic-forecasting platform:
- Hybrid LightGBM forecasting (global + per-route models) predicting travel times up to 4 h ahead — ~1.7 min MAE at 15-min lead, with conformal-calibrated uncertainty bands
- Real-time ingestion of ~27M rows/day into PostgreSQL/TimescaleDB
- Three GCP environments, GitHub Actions CI/CD with health-gated deploys
- LLM-generated traffic reporting via the Anthropic API
Live, public work (all running 24/7):
- bikeflow — open ML pipeline on live bike-share data: Airflow, nightly LightGBM forecasts with conformal bands, dbt + DuckDB marts, and a fact-checked AI daily brief
- transit-copilot — glass-box AI agent over Milan transit + the bikeflow warehouse: it writes and shows its own SQL; Claude tool use, MCP server, 12/12 evals, hard cost caps · ask it live
- four-step-model — the classical travel demand model for Milan on open data, with user-equilibrium assignment and a policy scenario · interactive maps
- routing-engine-osmnx — FastAPI shortest-path service on OSM networks · live demo
- Accessibility-using-Transit — transit accessibility with equity metrics · live maps for Torino, Milano, and Paris
Research: public transit equity at Institut Polytechnique de Paris — first-author papers at hEART 2022 and TRB 2023.
Python LightGBM TimescaleDB Airflow dbt DuckDB FastAPI GCP Anthropic API GitHub Actions