§ 04Projects & papers: systems, products, research

Projects & papers.

Eight pieces of work across the four lanes: production systems, shipped products, and the research they make possible.

Connectivity & walkability scoring

A project I led at KR&A: a connectivity and walkability score computed live for any coordinate in 38 countries across Europe, North America and APAC, with a precomputed 250 m grid as fallback, working with TBs of data and 100s of data sources. It began as one client's question and shipped as a production API, now integrated into a.s.r. real assets investment partners' PULSE platform.

PostGISS3Dijkstra & Network ScienceDistributed Compute

§ 04.02Developed at KR&AKR&A

Monthly house-price index · 13 EU countries

A multi-year study at KR&A where I was technical lead: a web-scraping pipeline across 13 EU countries feeding log-price hedonic regressions into monthly indices, tested as disaggregation indicators for Eurostat's quarterly HPIs; the work is officially published by Eurostat as a statistical working paper.

PythonScrapy/SeleniumPostGISMongoDBR

Side projects & research

§ 04.03MSc Thesis · BocconiRESEARCH

Gentrification agent-based model

My MSc Thesis research on gentrification, where I built an agent-based model of neighbourhood change: households interact with neighborhoods in the treated cities Amsterdam, Utrecht and Milan based on their own income, the affordability of the neighborhood and the attractiveness, based on elements such as streetview imagery 'beauty' classification, GTFS connectivity, greenery and neighborhood sentiment online. A two-tenure social-housing extension is in progress.

Python/GeoPandasLarge Vision ModelPostGISAgent-based Modeling

§ 04.04EngineeringPROJECT

Research pipelines as production systems

My research projects run on Airflow 3 data pipelines that are hosted in this project. It can scale from a laptop to a multi-machine CeleryExecutor cluster. This project was built with idempotency guards, custom operators and agentic monitoring.

AirflowCeleryDockerCIDuckDB

§ 04.05AI productPROJECT

LanguageBuddy: AI language tutor

A self-hosted AI language tutor for Dutch, Italian and Spanish, built on language learning research that includes a chat or voice-call LLM tutor and SM-2 repetition. Every mistake is captured into a spaced-repetition queue that drives the next day's exercises and uses a real-news scraper for reading learning. Follows the CEFR framework with 6,200+ vocabulary entries.

FastAPILLMTTSSQLiteDocker

§ 04.06NLP · sequence taggingPROJECT

SponsoredBye: sponsor-segment detection

A text-only sponsor-skipper for YouTube: sentence-T5 embeddings feed a BiLSTM sequence tagger that flags sponsored sentences and maps them back to timestamps, cutting segmentation error to 16% WindowDiff against a 99% naive baseline.

TensorFlowBiLSTMsentence-T5MongoDBHuggingface

§ 04.07Mobile MLPROJECT

FishFinder: photo-to-species ID

A Flutter app that identifies 63 Dutch fish species from a photo, fully on-device, and fills a Pokédex-style FishDex as you catch them. The training pipeline: ~3,000 hand-annotated photos masked with Segment Anything Model, then a fine-tuned ResNet50 (a 90 MB full model) compressed to an 8.8 MB TFLite model for on-device use.

Flutter/DartTFLiteResNet50Segment Anything Model (SAM)Firebase

§ 04.08BSc Thesis · UvARESEARCH

Predicting flooding risk from local features

My BSc thesis, looking at whether flood risk can be explained by local features instead of a black-box hydrodynamic simulation. Collected 33 features across ~45,000 European locations, such as ground imperviousness, ground type and distance to river, across 100GB+ of data and ran a Random Forest to find their relation. Findings: 97% on the binary 20-year flood question, with surrounding imperviousness and relative height doing most of the work.

scikit-learnRandom Forestraster dataGIS

§ 04Research: papers & methods

Papers.

Working papers, preprints and method write-ups on research in urban dynamics, network analysis, geospatial methods and data science.

2026-07-09

When Does Metro Infrastructure Capitalize into Property Prices? Phase-Decomposed Difference-in-Differences Evidence from Seven European CitiesWe study when, not merely whether, new metro lines capitalize into residential property prices. Seventeen staggered treated cohorts across seven European cities in five countries (Milano, Amsterdam, Copenhagen, Paris, Helsinki, Rennes, Roma) pool into a single phase-decomposed panel (n = 42,004), with the response decomposed into announcement, construction, opening, and maturity phases. The pooled cross-city average locates the largest response at maturity: prices step up by +9 to +12% (drop-Roma to full-sample, measured as the construction-to-maturity contrast) two or more years after opening, a step that is stable across the control ladder and positive under every leave-one-city-out. The step is a pooled average, however, not a within-city fact. City-by-year fixed effects collapse it to an insignificant -0.5 log points while leaving a +2.5 log-point step at opening, a within-city contrast that is itself entity-clustered ordering evidence, not significant under city-clustered bootstrap. Few-cluster significance of the maturity step is partition- and control-set-dependent: a restricted wild cluster bootstrap gives p = 0.036 clustered on the seven cities (level-only controls; p = 0.080 with the full control set) but p = 0.16 on the twenty-four cohorts. Roma, priced through the same OMI appraisal series as Milano, is itself a within-city null whose pooled contribution runs entirely through the common-year fixed effects, and it disciplines the upper-bound reading throughout. The defensible magnitudes are per-city: foremost Milano's within-ring +167 EUR/m² (≈ +5.6%, wild-bootstrap p = 0.004). We read the delayed-to-maturity step as a cross-city pattern worth testing on longer panels, not as a settled within-city effect.Download PDF ↓

IN PROGRESS

2026-08-14

Calibrating Free Postcode Boundaries from OpenStreetMap: A Transferable Seed-Density Accuracy CurvePostcode boundary polygons are unavailable free of charge for much of Europe: some countries publish authoritative layers, elsewhere they are sold or absent. Voronoi tessellation of OpenStreetMap address points is the natural estimator, but how many address points are needed, and whether the answer transfers across countries, has not been established. We calibrate a single OSM-Voronoi pipeline against national references in the Netherlands and Denmark (5,160 reference polygons), fit a seed-density-to-IoU curve, and test out-of-sample transfer to held-out Belgium (1,188 polygons). The asymptote is robust across functional forms (mean matched IoU saturates at 0.76–0.82), while the fitted 0.7-IoU threshold is form-sensitive (roughly 40–110 seeds); per-postcode scatter is wide (point-level R² of about 0.26), so the curve calibrates the population mean, not individual polygons. Belgium reaches mean matched IoU 0.618 at 81% coverage, near the curve. Applied to Italy, where no free authoritative intra-city postcode layer exists, the pipeline produces 4,209 estimated CAP polygons; Milan reaches mean IoU 0.783 against the community uMap reference (an earlier pipeline configuration), and the 2,903 matched single-CAP municipalities at median 19 seeds land where the curve predicts (mean IoU 0.500). A Swiss evaluation, where the asymptote falls to 0.644, marks the boundary of the regime.Download PDF ↓

WORKING-PAPER

→

2025-03-05

US vs EU: does training-data geography matter for autonomous-driving object detection?Does an object detector trained on US dashcam data still work on European streets? An original 2024 course project suggested catastrophic transfer failure, but was confounded: precision-only metrics, no held-out test set, a resolution mismatch between datasets, and models never trained in-domain. This controlled redo runs a 2×3 design ({zero-shot COCO, US-fine-tuned, EU-fine-tuned} × {US-test, EU-test}) with YOLOv3u and YOLOv8s on Udacity/CrowdAI (US) and KITTI (EU). On EU test data, EU fine-tuning gains +0.153 mAP@0.5:0.95 over zero-shot while US fine-tuning gains +0.001; the difference-in-differences gap is +0.077 ± 0.007 across three seeds. The pattern is mirror-symmetric, so the correct read is narrow fine-tune specialisation rather than a US-specific geographic bias, with one asymmetry that is genuinely geographic: bicycle detection collapses without European training data.Download PDF ↓

WORKING-PAPER

→

Social housing impact on the gentrification ABM

IN PROGRESS

Urban heat island research