Bring the workflow, artifact, or agent system that needs sharper boundaries.
Work that fits
Open-source inference. Evals. Fine-tuning. Agent orchestration. CAD/DFM workflows. Cognitive-security review.
The work starts with the thing that can fail: a tool call, drawing, operator decision, model output, data path, or review gate.
What I build
Task suites, trace capture, verifier checks, replayable runs, failure taxonomies, and review bands.
For local AI work, I help choose models, prepare datasets, run evals, and deploy useful workflows without treating the model API as the whole system.
Proof paths
Applied Medical: tooling, fixtures, and a manufacturing process fix worth roughly $100k/month.
T-UEBA: constrained graph ML, calibration, active learning, and analyst evidence paths.
Agent environments: state, tools, traces, verifiers, and feedback loops before reward theater.
Contact
Email malachi@outlook.com.
Send the workflow, current system, costly failure mode, and what already exists: data, model, eval, deployment surface, or customer process.