About
I build production machine-learning and data systems: forecasting models,
real-time pipelines, and the cloud platforms that run them. At CitiEU — CitiME in
Milan I built and operate a traffic-forecasting platform end to end —
a LightGBM system predicting travel times up to 4 hours ahead, a real-time
ingestion layer processing ~27M rows/day into TimescaleDB, and LLM-generated
reporting — deployed on GCP with CI/CD.
Before that, I researched public transit equity at Institut Polytechnique
de Paris, published as first author at hEART 2022 and TRB 2023.
Experience
- Designed and operate a hybrid ML forecasting system — a global
LightGBM model trained on pooled route history (~1.5–3M observations) blended
with per-route specialist models by recent accuracy — predicting route travel
times 15 min to 4 h ahead: ~1.7 min MAE at 15-minute lead,
16–50% better than a typical-day baseline, with quantile +
conformal-calibrated P15–P85 uncertainty bands (~80–85% empirical coverage).
- Own the full ML lifecycle in production: nightly automated retraining,
every forecast graded continuously against actuals per route, lead time, and model
variant, and cold-start handling for new routes through the global model.
- Built the platform's real-time ingestion layer: ~27M rows/day of
per-minute traffic telemetry into PostgreSQL/TimescaleDB — hypertables with
columnar compression (371 GB logical stored in 137 GB) and continuous aggregates
that keep month-scale analytics off the raw data.
- Run three GCP environments (production, staging, delivery) with
GitHub Actions CI/CD: health-gated auto-deploys, PR tests against a real TimescaleDB
service container (~75 Node test files), Secret Manager for all credentials, and
idempotent SQL migrations on every deploy.
- Integrated the Anthropic API for AI-generated traffic incident and
trend reports, and built a bash multi-agent audit framework (Claude + Codex) running
8+ specialised review roles per script for code, UI/UX, and performance audits.
- Migrated a legacy Cube Voyager transport model to CubePy, cutting end-to-end
runtime by 40% on the Abu Dhabi Mobility model, and trained
50+ modellers on the new stack.
- Built a Python pipeline ingesting multi-city GTFS feeds into MongoDB and computing
accessibility metrics (Lorenz curves, Gini indices) entirely from open data.
- Contributed to the French ANR MuTAS project (ANR-21-CE22-0025-01); output published
at hEART 2022 and TRB 2023 as first author.
Projects
IntelligentMobilityHub
Production ML & data platform · CitiEU — CitiME
Traffic-forecasting platform run in production: per-minute telemetry (~27M rows/day)
ingested into TimescaleDB, a FastAPI/LightGBM service forecasting travel times up to
4 h ahead with calibrated uncertainty bands, MapLibre GL live map, and
Anthropic-API-generated reports — deployed on GCP with health-gated
GitHub Actions CI/CD.
LightGBMTimescaleDBFastAPIGCPGitHub Actions
bikeflow
Personal project · live dashboard →
Open, end-to-end data pipeline collecting real-time bike-share availability
for Milan and Paris (GBFS): Dockerized collector on a GCP VM, Airflow 3
orchestration (compaction, GCS backups, freshness alarms), CI, and an
hourly-published public dashboard. Built as the open counterpart to my
production work — LightGBM forecasting layer in progress.
AirflowDockerGCPParquetGitHub Actions
Abu Dhabi Mobility Model
Migration & performance · CitiEU — CitiME
Ported Voyager scripts to CubePy and resolved an ArcGIS Engine dependency blocking
the model on newer environments; cut end-to-end runtime by 40% and trained 50+
modellers on the migrated stack.
CubePyCube VoyagerPython
Public Transit Equity Analysis
Research · IP Paris
Stored and queried multi-city GTFS feeds in MongoDB; computed Gini- and Lorenz-based
equity scores at the transit-line level using open data only. Output: two first-author
conference papers.
PythonMongoDBGTFS
Publications
Badeanlou, A., Araldo, A., Diana, M. · hEART Conference 2022
A fully automated methodology to measure transportation accessibility inequity using
Lorenz curves and Gini indices, relying solely on open data — demonstrated on four
cities and compared with existing approaches.
Badeanlou, A., Araldo, A., Diana, M., Gauthier, V. · Transportation Research Board (TRB) 2023
An equity scoring system for public transit lines aimed at reducing car dependency in
underserved areas — measuring and improving accessibility equity across seven cities
from open data, to guide large-scale transit optimization.
Skills
- Machine Learning & AI
- LightGBMQuantile regressionConformal calibrationFeature engineeringModel monitoringAnthropic APIClaude Code / MCP
- Data Engineering
- PostgreSQLTimescaleDBReal-time ingestionETL/ELTREST APIsMongoDBSQLite
- Cloud & DevOps
- GCP Compute EngineSecret ManagerBigQueryGitHub Actions CI/CDpm2Bash
- Programming
- PythonpandasNumPyGeoPandasSQLJavaScriptNode.jsFastAPI
- Geospatial & Visualisation
- MapLibre GLQGISArcGISGTFSDashboards
- Transport Modelling (domain)
- OpenPaths CUBECubePyAimsunFour-step modelling
Education & Honors
Politecnico di Torino
M.Sc. Civil Engineering — Infrastructures & Transportation Systems · Turin, Italy · 2019 – 2022
Sharif University of Technology
B.Sc. Civil Engineering · Tehran, Iran · 2015 – 2019
- Rapid promotion — promoted to Abu Dhabi Mobility Transport Modeller within 12 months (typical track: 3 years).
- Full M.Sc. scholarship — awarded for the duration of the Politecnico di Torino programme.
- Research grants — funding secured from three independent sources for the Paris-based master's thesis.
- Conference presentation — presented first-author work at hEART 2022.