I turn data and AI ideas into production-ready products:
9+ years experience with geospatial data, LLM and ML systems at scale.
Business Challenge
How to use hyped AI tools as an individual, a team, or a whole company?
Solution
2-day hands-on workshops on working effectively with coding agents: the mechanics underneath, the role shift on top, the craft in between.
Win
Participants confidently use coding agents like Claude Code. The way of working outlasts the tool.
Business Challenge
Efficiently access the best up-to-date climate reanalysis.
Solution
Analysis-ready cloud-optimised (ARCO) ERA5 on Earthmover Arraylake, dual-chunked for map and time-series access, with Icechunk ACID transactions and an SLA on freshness updating daily.
Win
Customers can subscribe to a ready-to-use ERA5 dataset. Time-series and map queries execute in under 3 seconds.
Business Challenge
How to speed up the candidate review load of recruiters and hiring managers at scale?
Solution
Agent narrowing 20M profiles to 1,000 via embeddings retrieval, down to ~200 prescreened by a fine-tuned LLM, ending at ~10 qualified and interested applicants per job posting.
Win
Reduced time spent on shortlisting for hiring managers to 10 minutes.
Business Challenge
Maximize revenue per impression by ranking ads by relevance for the user.
Solution
Upgraded from logistic regression to a deep neural network enabling ad-user personalisation.
Win
Improved click-through-rate by 10–15% and revenue-per-impression by 2% in production.
Business Challenge
Run a Kaggle-style challenge for sub-seasonal climate forecasts with low entry barriers
Solution
Curated training datasets and a verification pipeline based on git-lfs and renku notebooks for forecast submissions automatically evaluated on standardised metrics.
Collaboration
Swiss Data Science Center (SDSC), World Meteorological Organization (WMO) and European Centre for Medium-Range Weather Forecasts (ECMWF).
Aaron Spring
