Heimdall organises machine learning around two distinct starting points. The first is the classic data path: you bring CSV files or connect a database, ingest and version them across bronze, silver and gold layers in Lake, explore and profile tables with Python notebooks in Lab, then publish gold tables that the ML and Forecast products train on. ML covers classification and regression, while Forecast produces time-series predictions for demand and capacity planning with explanations you can inspect. Every model finishes as a monitored REST endpoint with usage metering and a health check.
The second path is for teams who have no training data yet. Loop is an online-learning product whose models keep adapting to fresh live feedback rather than a fixed training set, which suits recommendation and ranking style problems. Resolve takes unstructured text and turns it into a category or a score using answers you define yourself — no labelled history required — which makes it useful for routing incoming emails or grading urgency. Both paths still end in the same deployable, monitored API surface.
Because the platform bundles cataloging, notebooks, modelling and deployment, it is a fit for small data teams that want production machine learning without assembling separate warehouse, notebook and MLOps vendors. Access starts with a free account and a self-serve workspace; larger deployments are handled through a conversation with the team, so published prices are not listed publicly.
