One small, honest ML system per day, built end to end on Hopsworks.
Estimate an asteroid's size (hence impact threat) from its Gaia DR3 reflectance spectrum, beating the blind constant-albedo guess (size error x1.34 -> x1.13). An FTI ML system on Hopsworks.
Predict GitHub repo abandonment from its README alone. 3600 labelled repos, FTI pipelines on Hopsworks. Turns out hype doesn't predict it, README thinness does.
Every aircraft over Europe, live: trajectory prediction that must beat physics, GPS-jamming detection, learned airspace normalcy. Real-time FTI on Hopsworks.
Predict bioactivity of never-tested natural products from molecular structure. LOTUS + ChEMBL, multi-task QSAR with calibrated applicability domain, AMR-led. FTI on Hopsworks.
Ground-level air quality where no sensor reaches: satellite, stations and weather fused in the feature store
Who is hiding on the ocean: vessels scored live by how much their behaviour resembles the sanctioned shadow fleet, plus the network they meet in the dark. Real-time FTI on Hopsworks.
Photo in, country out. Frozen CLIP + a head trained on OSV5M that must beat zero-shot. Full FTI ML system on Hopsworks: the feature store holds vectors, never images. #004 of awesome-ml-systems
Flag phishing domains at TLS-issuance time, straight off the Certificate Transparency firehose, before the site serves a byte. Hostname classifier over 15k balanced hosts (holdout ROC-AUC 0.78, 0.88 precision at the blind baseline's recall). A self-retraining FTI ML system on Hopsworks.
The machine calls your click before you make it. Online per-user preference learning as a game, on Hopsworks.
Add a description, image, and links to the awesome-ml-systems topic page so that developers can more easily learn about it.
To associate your repository with the awesome-ml-systems topic, visit your repo's landing page and select "manage topics."