Retrieval evaluation
A local document retrieval toolkit with an evaluation harness for source retrieval, citations and latency.
Put AI to work.Know what it’s doing.
Helping businesses put AI to work: finding where it saves real time, building the tools and automations, and setting them up to stay reliable and safe.
How I can help
Practical AI, set up to stay safe.
For teams that want AI to take real work off their plate without creating new risks.
Find the time savings
A close look at how the team actually works, then the jobs AI can take on: sorting email and documents, research, reporting and customer follow-up.
Build and connect it
Assistants, automations and data pipelines, set up inside the tools already in use, so nobody has to change how they work overnight.
Keep it safe
Clear limits on what AI should never touch, checks on what it produces, and proper handling of customer data, informed by tracking how AI systems fail in public at Paperclip Index.
Lead project
Tracking the gap between what AI is asked to do and what it does.
Reports of AI systems causing harm or crossing the limits people set are scattered across lab disclosures, news and court filings. The index rates each one on fixed scales and turns those ratings into a reading anyone can trace back to its sources.
Methodology as a spec
Any change to a weight, band or eligibility rule ships as a new version, with a changelog naming every record it touched.
Evidence before ratings
Each record separates documented harm from alleged, disputed and unknown claims. Discovery-only sites can lead to a story but never serve as its evidence.
Pipelines that do the legwork
Automated pipelines find new reports, keep the backlog sorted and keep every page consistent with the site’s design. That frees my time for the editorial work: I set every rating and edit every summary.
Other work
Retrieval, evaluation and tools.
Projects in retrieval evaluation, on-device ML and data tooling.
3.0.0 Preview 1
A browser extension that runs an embedding model locally to find saved pages by meaning.
Data tooling
Dataset preparation, label review and classification baselines for messy real-world text.
Self-hosted · Request access
Image tournaments for Discord communities, from submissions and reaction votes to the final bracket.
Background
An engineer who’s run a business.
For three years I co-owned and ran a retail business: managing a team of more than ten, plus the books, marketing and daily operations. I know firsthand how tight time and budgets get.
Then I joined Fuel Me, a startup, as an electrical engineer, leading prototyping and field deployment of a real-time tracking system and working closely with large retail partners to test it in the field and roll it out.
- Education
- B.S. Electrical Engineering, University of Wisconsin–Madison
- Certification
- AWS Certified Machine Learning Engineer – Associate
- Previously
- Electrical Engineer, Fuel Me · 2022–2023Co-owner, Rainbow Ceramics · 2018–2021
Contact
Get in touch.
Wondering where AI fits in your business, or what could go wrong when it does? Send a note about what you’re working on for a straight answer on whether AI can help.