Freelance ML engineer, AI engineer and geospatial data engineer based in Hamburg. I build:
Business Challenge
Access the best climate reanalysis based on the most recent available data for maps and time-series use-cases.
Solution
Built an analysis-ready cloud-optimised (ARCO) ERA5 dataset on Earthmover Arraylake with dual chunking optimized for maps and time-series applications, leveraging Icechunk for ACID transactions and an attached SLA covering data freshness and availability.
Business Challenge
How to speed up recruiter candidate review load at scale?
Solution
Recruiting agent narrowing 20M profiles to 1,000 candidates via embeddings retrieval, down to ~200 prescreened candidates with a fine-tuned LLM, returning ~10 qualified & interested applicants via personalised outreach per job posting.
Business Challenge
Maximize revenue per impression by selecting the ad most relevant for the user.
Solution
Improved CTR and revenue-per-impression by 2% by upgrading from logistic regression to deep neural network enabling ad-user personalisation.
Business Challenge
Run a Kaggle-style challenge for sub-seasonal climate forecasts with low entry barriers
Solution
Curated training datasets and a verification pipeline on git & notebook-based renkulab.io for clean-data ML forecast submissions with standardised metrics.
Aaron Spring
