Unsupervised

AI systems for data analysis.

Work

SemLang

ReleaseMIT

A semantic analytics language that compiles to Malloy, so agents query business concepts instead of guessing at raw tables.

HHS Medicaid analysis: a full run, published

Artifact

An overnight run over 194 million Medicaid claims: drivers, segment patterns, caveats, and open questions, with backing SQL.

Evals

ReleaseMIT

Declarative, container-based agent evaluations that produce reproducible, attestable results.

Structure makes frontier models more expensive and open-weights models cheaper

Evaluation

The same 104-task agent benchmark on a frontier model and an open-weights model, raw and under structured workflows, with the bill for each.

Workflows written by a frontier model made an open model twice as fast

Evaluation

An open-weights model ran 104 one-day agent tasks raw, then again executing workflows a frontier model had written. An early readout, with the work shown.

All writing →Agent tooling ↗

Enterprise

Finder for Teams finds statistically significant patterns across related tables, ranked by KPI impact. Read customer stories, or get in touch.

Updates