Stephen Solka trashhalo

Stephen Solka

I build AI models, evaluations, and the systems around them—from specialist retrieval to models that make decisions.

DecisionBench — Evaluating decision models on practical tasks.
Bosun — Compact models for routing, classification, scoring, and abstention.
Dinghy Law — Specialist embeddings for legal retrieval.
LangSet — Training language models to predict future latent states.

Obsidian Webhooks — A self-hosted Go server and Obsidian plugin that turn webhooks into note updates.
ML Recipes — Runnable ML tutorials using real datasets.