Randal S. Olson, Ph.D.

• Owned product and engineering from day one and wrote the production code.
• AI evaluation platform: built to turn domain experts' no-code judgments into production LLM evals, deployed as live API endpoints that engineering teams use to score AI outputs at scale.
• Agent workflow platform: rebuilt the eval platform into one where agents run AI workflows that automated verifiers check, shipped with a Python CLI, an MCP server, agent skills, and production infrastructure. Built verifier-driven content QC on it for a major edtech customer. Both products reached paying customers.

Senior Data Scientist, Jun 2024 to Mar 2025; Head of AI Strategy, Mar 2025 to Aug 2025

• Speech AI for early readers: as Senior Data Scientist, built an app that checks young students' phoneme and letter-name pronunciation in real classrooms, plus a custom eval on 500+ labeled classroom recordings. The eval cut iteration from days to hours and showed a hyped model upgrade was not worth migrating to.
• Solution architecture: as the technical lead in sales, scoped AI solutions for prospective clients in edtech, biotech, finance, and enterprise software, from their highest-ROI opportunities to the systems the team would build.

Short engagements designing and building custom AI systems, plus AI strategy, for select clients.

• Founding member of the AI research team.
• Optimization systems: built two multi-objective optimization systems from scratch for the wet lab teams: a genetic algorithm that optimized E. coli growth media for drug production, which contributed to an Absci patent filing (opens in new tab), and a model that balanced antibody binding affinity and naturalness, which led to Bachas et al. (2022) (opens in new tab).
• Data pipelines: built Python ETL between the wet labs and the AI team, cutting data transfer from days to minutes.
• MLOps: built the infrastructure the team used to pretrain, fine-tune, and evaluate LLMs for drug discovery.

Lead Data Scientist, 2018 to 2020; Chief Data Scientist, 2020 to 2021; Scientific Advisory Board, 2021 to 2022

• Founding data scientist: built and led the data science team that brought epigenetic testing to life insurance risk assessment.
• Production ML: built statistical ML models that predict health and lifestyle traits such as smoking status and alcohol use from saliva-based DNA methylation data, and took them from prototype to production in a customer-facing platform.
• End-to-end data stack: open source DNA methylation processing (methylprep (opens in new tab)), a serverless AWS pipeline, and a proprietary research database.

Prof. Jason H. Moore's lab, Institute for Biomedical Informatics

• TPOT: created one of the first AutoML libraries (opens in new tab), which grew into a standalone open source project (intro post).
• Research to software: turned the lab's research code into maintained Python packages, including PMLB (opens in new tab), scikit-rebate (opens in new tab), and scikit-mdr (opens in new tab).

Built a high-performance C++ platform for simulating how intelligent behavior evolves, and published on the evolution of cooperation and intelligence.

Built software for the Electronic Warfare division as part of a larger engineering team.