Senior Data Engineer at Chimnie · Bengaluru

Shubham Agarwal

Geospatial data engineer. I turn aerial imagery, LiDAR and satellite data into national-scale datasets, most recently a footprint for every building in Great Britain. I got here via football.

See the work Email me LinkedIn GitHub
Shubham Agarwal standing by the sea pictured: touching grass
01

My story, in halves

Hi there, and welcome.

Kick-off2018 · sports

My data journey started with sports. I've always been drawn to the moments of magic that happen in a stadium, and by college I knew I wanted to work in the industry behind them. That led to my first internship in 2018, where I learned how sports businesses actually run and, more lastingly, got hooked on data science.

First half2019–21 · the startup

With minimal experience, maximal ambition and a nudge from my MBA, I co-founded a sports analytics firm in 2019 to fill gaps I saw in Indian sports analytics. We were bootstrapped, signed academies across Bengaluru, and quietly hoped to help India reach a FIFA World Cup one day. Then the pandemic arrived, and it is hard to analyse sports when no sports are being played. We shut down. It was a tough stretch, and also the one where I grew the most technically.

Second half2021–24 · climate tech

The pandemic also changed what I wanted to work on. Climate change felt like the next large, slow-moving risk, and I wanted my skills pointed at it. From 2021 I moved deliberately into climate tech, focusing on solar energy and carbon-emissions monitoring, where my interests and my potential impact overlap the most.

Extra time2024– · maps

In August 2024 I joined Chimnie, a UK property-data company, and went from analysing footballers to mapping roofs. I led Polygons, a model that reads aerial photos and LiDAR height data to outline every building in Great Britain: 24.2 million of them, each linked to its addresses.

Full time?not yet

Today I'm looking to own bigger problems end to end: the data, the model, and the people who rely on the result. If you're working on energy, mapping or climate and could use that, I'd be glad to talk. Cheers!

02

Work

Eight seasons, three leagues.

Senior Data Engineer

Chimnie · UK property dataAug 2024 – now
  • Led Polygons, a national roof-footprint dataset: 24.2 million buildings across England, Scotland and Wales, linked to all 35.4 million addresses. It finds 98% of the addressed buildings in the national reference map, a dataset licensed at £5.4 million a year.
  • Built the model: a ConvNeXt segmentation network fusing 25 cm aerial imagery with LiDAR height, pre-trained on a multi-country building dataset and fine-tuned on UK labels. Held-out building F1 of 0.90.
  • Ran national inference on GCP: a Vertex AI GPU fleet over 9,947 tiles, post-processed and published to cloud storage. Launch to release in 3 days.
  • Designed the labelling system (label contract, custom annotation app, vision-model drafts behind a human approval gate) and reviewed every training label. Ran the build with a team of AI coding agents behind review gates.
  • Delivered regional client datasets, such as Greater Nottingham (260,828 buildings, 402,827 addresses), as GeoParquet and GeoPackage.
Computer visionPyTorchLiDARVertex AIGeoParquetDuckDB

Senior Data Scientist

Blue Sky Analytics · climate and energy dataOct 2023 – Jul 2024
  • Ran a solar-asset pilot for an Indian renewable-energy company: 3 solar farms, about 3,000 ha and 1,000 MW. Built panel segmentation and detection, and annual irradiation and energy-yield forecasts, from raster and LiDAR data.
  • Pitched data products to win pilots with top-3 companies in sustainable palm oil and petrochemicals.

Data Scientist

Blue Sky AnalyticsNov 2021 – Sep 2023
  • Deployed a containerised U-Net on AWS Lambda that segments smoke plumes in satellite images of fossil-fuel power plants, with a regression layer estimating daily CO₂ emissions.
  • Built climate-risk and emissions datasets from remote sensing and IoT data with Python, AWS and Docker, served through the company's APIs.
SolarEmissionsU-NetAWS LambdaGoogle Earth Engine

Associate Data Scientist

SportsKPIMar 2021 – Aug 2021
  • Built Python pipelines that clean and load match data into MySQL on AWS.
  • Helped the AIFF select India's U-23 football squad with insights from a state-level tournament.
  • Created "True Raider Impact", a regression-spline metric that Pro Kabaddi teams used at the auction. Recruited and mentored the data-engineering interns.
Earlier: founder, and the internship that started it

Founder and CEO

Statosphere TechnologyMay 2019 – Jan 2021
  • Signed partnerships with football and basketball academies in Bengaluru and gave their coaches player-level analytics for match and training decisions.

Data Analyst Intern

InTheGame AnalyticsJun 2018 – Dec 2018
  • Built an "expected goals" model with logistic regression and football dashboards in Tableau and R.
  • Made daily content that raised Facebook interactions by 12%.
Sports analyticsMySQLRegression splinesTableau

Education

and a couple of certificates
  • MBA, ESG Paris–India (2018–2020) · BA, Christ University, Bengaluru (2015–2018)
  • IBM Product Manager and Agile · Imperial College London, Mathematics for Machine Learning
03

Projects

Side quests, mostly sports and sunshine.

Map of roofs in San Francisco's Outer Sunset coloured by yearly sunlight
Solar · geospatial

How much sun hits every roof?

Rooftop solar potential for 3,735 buildings in San Francisco's Outer Sunset, from open LiDAR, building footprints and a typical weather year.

  • Roof slope, aspect and shading from a 0.5 m LiDAR surface model, then pvlib irradiance for every roof pixel.
  • Usable area, system size, yearly kWh and CO₂ avoided per building, checked against PVGIS and Google Project Sunroof.
Explore the map →Code →
NBA draft board ranking the 2026 class
Product · NBA

NBA Draft Projections V4

Every draft class since 2010, ranked by what pre-draft stats say, next to where teams actually picked.

  • Peak and Immediate Impact Ratings with ranges, the stats driving each projection, and ceiling and floor comps.
  • Tested on each class before it was seen: stats plus draft slot rank early-career impact better than the teams did (0.48 vs 0.42 Spearman) and match them on peak, with a home-made RAPM as a second target.
Draft board →Code →V2 app →
Dragon Ball Z powering up
Product · anime

Anime Scout

Tell it a few anime you've watched; it tells you what to watch next.

  • A hybrid recommender over 12,037 anime and 23.7M user ratings: what fans also liked, plus genres, themes and synopses.
  • AniMap, a treemap of 20 taste groups to wander through.
Find your next anime →Code →
Chart from the Bengaluru air-quality data story
Data story · air quality

The COVID lockdowns were a breath of fresh air for Bengaluru

If only they weren't laced with anxiety, death and loneliness.

  • ANOVA tests for a real change in air quality before and after travel restrictions.
  • An LSTM forecast trained on pre- and post-lockdown data, compared side by side.
Code →Blog →
Chart from the NBA Draft 2021 analysis
Data story · NBA

NBA Draft 2021: an analysis

Winners, losers, and a machine-learning look at how to draft well.

  • Scraped Tankathon, Basketball Reference and nba.com to find what makes a good pick.
  • Averaged regression models to rank the class, and classification to find player archetypes.
Code →Blog →
04

Skills

The kit bag.

Geospatial and Earth observation

GDALRasterIOGeoPandasshapelyPostGISGoogle Earth EngineQGISGeoParquetLiDAR / nDSMOptical and SAR imagery

Machine learning and computer vision

PyTorchtimmscikit-learnTensorFlow / KerasU-NetConvNeXtSegFormerYOLOSAMNLP

Data and cloud

PythonSQLpandasNumPyPyArrowDuckDBGCP: Vertex AI, GCS, Cloud RunAWS: Lambda, S3, EC2, SageMakerDockerGit

Statistics and forecasting

Hypothesis testingANOVAA/B testingBayesian methodsARIMA / SARIMAXLSTMRegression splinesClustering and PCA

How I ship

AI coding agents (Claude Code, Codex)Review gatesStreamlitTableauWeb scrapingPitching to executives

Domains and languages

Renewable energyEmissions and climate riskProperty dataSportsEnglishHindiKannada