PROJECTS
What I build and run
Three research systems I work on at That Recommender Systems Lab, and the infrastructure I use to run my own projects.
LEAD ENGINEER · THAT RECOMMENDER SYSTEMS LAB
Corgi: a community-governed Bluesky recommender
A live, open-source recommender that makes ranking policy inspectable, changeable, and auditable. Every scored post stores its raw signals and policy weights, so each placement can be reconstructed. Pilot participants vote on ranking weights. The production feed is live; governance is a small pilot.
View the code on GitHubLEAD ENGINEER · THAT RECOMMENDER SYSTEMS LAB
CARAWAI: consent-first research capture
A local-first browser research system that captures analytic activity with consent and turns it into evidence-cited workflow traces and summaries. I expanded an initial proof of concept into an end-to-end platform: capture and provenance, cross-session retrieval, policy-gated privacy controls, and fail-closed participant safety, built to collect as little as possible.
CONTRIBUTOR · NSF-BACKED, FIVE UNIVERSITIES
POPROX: live news-recommendation experiments
A platform that runs live news-recommendation studies with Associated Press content. I built foundations for multi-source storage and ingestion, and now work on recommendation contracts, outbound data minimization, and production hardening. That includes a deliberately staged ProPublica integration that stays isolated from production flows until the downstream path is validated.
PERSONAL INFRASTRUCTURE
An engineering control plane for agent-driven delivery
I built this after concurrent workstreams repeatedly collided or finished without a trustworthy history. Agents handle implementation and integration, while deterministic infrastructure controls task ownership, dependencies, isolated workspaces, review, and recovery. Every change leaves an auditable record. I use it to run my own software projects.
Detailed write-ups are in progress.