ML Systems
Research engineering across data, evaluation, reliability, and production feedback loops.
Zach Olivier works on machine learning systems where modeling quality depends on the surrounding infrastructure: data preparation, experiment tracking, evaluation design, observability, retrieval, and deployment feedback. The work is closest to research engineering and applied ML engineering, with an emphasis on systems that make model behavior measurable.
What I Build
- Evaluation systems for benchmark design, verifier scoring, regression checks, and distribution-shift analysis.
- Post-training workflows for supervised fine-tuning, preference data, best-of-N selection, and experiment reporting.
- Retrieval and knowledge systems for local-first paper ingestion, notes, concept graphs, and research agents.
- Reliability modeling tools for survival analysis, field-to-test translation, telemetry retention, and failure analysis.
- Applied modeling labs for tabular foundation models, industrial anomaly detection, and uncertainty-aware baselines.
Representative Projects
- Concensus SFT: a scientific QA supervised fine-tuning pipeline with data preparation, training artifacts, and loss diagnostics.
- Second brain knowledge pipeline: a local-first research system for papers, reviewed article nodes, and concept graphs.
- Industrial anomaly detection: a PatchCore-style visual inspection lab with retrieval over memory-bank features.
- Tabular foundation model lab: TabPFN and TabICL experiments against classical baselines with runtime and uncertainty notes.
- Reliability modeling templates: practical analysis patterns for survival, failure, and field reliability data.
Research Areas
Current interests include post-training evaluation, ML telemetry, retrieval-augmented research tools, tabular foundation models, industrial computer vision, reliability and survival modeling, and artificial-life experiments that stress test search, emergence, and measurement.