Now

September 2026. Current technical focus for Zach Olivier.

Working

• Vector Bucket: music similarity, semantic audio search, and interactive embedding maps.
• Conformal Lab: coverage, prediction sets, calibration, and uncertainty diagnostics.
• Post-training harnesses for SFT, DPO, verifier scoring, and best-of-N selection.
• Telemetry retention for ML systems: logging budgets, summary interfaces, and failure analysis.
• ALife experiments: replicators, program soups, mutation search, and population traces.
• Tabular models: foundation-model baselines, uncertainty, runtime, and training dynamics.

Planning

Looped LLM Lab: controlled experiments on recurrent transformer depth, adaptive halting, and sparse experts. The research plan and hardware diagnostic are in place; model implementation and training runs are still ahead.

The reading queue includes Mixture-of-Depths, Reasoning with Latent Thoughts, and When Does Recurrence Become an Algorithm? Related inference-time work includes Recirculation.

For tabular models, I am keeping a reading note on XGBoost vector-leaf models.

Testing

• Evaluation signals under distribution shift.
• Benchmark design that separates model behavior from dataset fit.
• Which project repos need hosted demos, benchmark results, or just a clean README.

Open To

• ML engineering and research engineering roles across data, models, evaluation, and production feedback.
• Applied research work at labs, scientific organizations, or product teams with a concrete dataset, experiment, or system to ship.