Director of Engineering for CiviForm at Exygy. CiviForm is open source. It is built by Exygy and Google.org with the City of Seattle and community contributors, and it exists so that applying for one government benefit does not mean answering the same questions again for the next one.
I lead engineering for public-interest systems, and I still build them. Most of my public repositories are measurement tools: validators, scorecards, and evaluation harnesses pointed at real transit feeds, hospital price files, health-plan APIs, credential records, college disclosures, and California energy filings.
chelseakr.com · LinkedIn · Davis, California
Every public repository here was built since June 2026, and AI agents are part of how I work. I choose the architecture, write the acceptance criteria, review the output, and decide what is ready to release. That is why this much exists in this little time.
It is also why so much of it is gates. The failure mode of working this way is a check that cannot fail and a green build that means nothing, so each project is built so that it can fail, and when it does the failure is written down where you reach it before you reach the claim.
These are independent personal projects with no proprietary or client material in them. They are not CiviForm or Exygy work, and nothing in them speaks for either.
These projects have very few stars. What they have instead is a record of being corrected in public.
A project's scope. Jannis (derhuerst), a longtime open transit-data maintainer, argued that GTFS Scorecard's scoring belonged inside MobilityData's canonical validator rather than in one more dashboard. He was substantially right. I named in the thread the tools I had duplicated, declined to push subjective letter grades into an official project where they would read as guidance, and narrowed this one to the handoff nobody else covers.
A scoring rule. The person who produces the MRC de Joliette feed
pushed back on a recommendation to
populate trip_headsign on loop routes. They were right, and the rule now credits that case
instead of flagging it.
Upstream, MobilityData merged a specification example fix to the Transit Operational Data Standard and an awesome-transit listing. A conformance-language clarification, a second awesome-transit listing, and a Transitland feed-archival PR are open. Five small contributions, two of them merged. That is the whole claim.
There are more repositories here than anyone wants to browse. They fall into six groups, and each group has one worth reading first.
- Transit data. Start with
gtfs-scorecard. Thentods-validate,fare-policy-assistant,transit-delivery-atlas. - California filings and public records. Start with
qfer-preflight. Thenpower-content-check,ca-tariff-parse,ceqa-preflight,permit-bearings,perimeter,wildfire-service-territory-overlap. - Standards conformance. Start with
ctdl-validate. Thenoscal-validate,fhir-scorecard,mrf-honest,ctdl-validate-jvm. - Education and workforce. Start with
afterward. Thendisclosed,homeroom,chalkline. - Evaluation gates and receipts. Start with
plumbline. Thengauntlet,cairn,sprout,contextsafe,outcome-receipts,obligation-receipts,exitdrill. - Community and personal tools. Start with
habitable. Thenledger,swelter,nearmiss,id-churn-sentinel,davis-bike-hazard-map,queer-the-stacks,encore,family-greenhouse,olive-bark-logger,lavender-rotation.
Most are pre-1.0, and I would rather you take a repository's own word for its maturity than mine.
- GTFS Scorecard (live at gtfsscorecard.org) grades more than 2,100 curated transit feed records daily in plain language. Correctness findings come from MobilityData's canonical validator, not a competing one. Because it names public agencies it publishes a listing policy, and its remediation handoff is badged Pilot because that pilot has not recruited a participant yet.
- tods-validate checks Transit Operational Data Standard feeds, the crew runs and vehicle assignments GTFS does not cover, against TODS v2.1.0, with rule IDs that are never renumbered and a GitHub Action so an agency gates a bad feed before publishing. Reading that spec closely enough to write the rules is what produced the two upstream TODS pull requests above.
- ctdl-validate structurally validates CTDL JSON-LD, the national credential-data standard, citing the published schema behind every finding, as a CLI, a GitHub Action, and a browser playground that uploads nothing. Run against 120 documents sampled from the public Credential Registry it returned forty ERROR findings, every one tracing to an inconsistency inside CTDL's own schema encoding rather than to a publisher's mistake. That is a finding about the tool as much as about the corpus, and the write-up leads with it.
- afterward joins 3,266 California training programs reported under WIOA to the state's own ten-year projection for the occupation each leads to, in English and Spanish, with no account and no tracking. A suppressed or unreported outcome never renders as zero, and how much the join actually covers is a published output rather than a footnote.
- fare-policy-assistant answers rider questions about reduced-fare policy over a corpus covering eighteen California transit agencies. The assistant exists so that the public evaluation harness has something to evaluate: 385 cases, versioned prompts, a committed regression baseline, merge-blocking refusal and grounding gates, and a standing record of what it has not measured, including that none of its 28 Spanish answers has been rated.
- habitable makes habitability evidence for tenant unions tamper-evident with content hashes and RFC 3161 timestamps, then syncs peer to peer under end-to-end encryption so there is no central holder to subpoena. Its own README says not to rely on it for real legal matters yet, and I agree with it.
A tool that grades other people's work has no business shipping a check that cannot fail. Hunting those down in my own repositories is a standing part of how I build.
plumbline is a fail-closed audit harness. A target that
returned 174 empty responses scored 1.0000 on groundedness, privacy, representational harms,
fairness and cross-language, and the gate exited PASS on that alone, because each of those checks
is phrased as the absence of a bad thing and silence satisfies all of them.
The fix
splits the two kinds of check apart, and a test now fails the build if a silent target ever passes
again.
That same shape turned up again and again elsewhere, every instance green at the time.
ceqa-preflight printed four
PASS lines over a package whose PDFs had all timed out, one of them affirmatively clearing a
document it had never opened. A weekly monitor in
id-churn-sentinel went
green four weeks running while zero of its 152 sources were eligible to check, because observing
nothing emitted the same signals as observing no change.
Before Exygy I was a Senior Director of Engineering at Coforma, one of three in a 50-person engineering organization, leading a 22-person reporting structure and owning the company-wide healthcare engineering portfolio. Nine engineers moved into senior or leadership roles under me, four of them to Director or Principal.
Before that I was engineering lead and principal engineer for MyCareer.NJ.gov, New Jersey's statewide workforce platform, where I set architecture across three production codebases, ran a zero-downtime GCP to AWS migration, shipped full English and Spanish parity, and did the 2023 CTDL data modeling that began New Jersey's migration of its training-program registry into the national Credential Engine ecosystem. Earlier I built public systems for the California Energy Commission, the California Public Utilities Commission, the California Department of Social Services, and UC Berkeley's Graduate School of Education. I am a registered member of Credential Engine's CTDL Advisory Group.
The domains I keep returning to are public health, workforce and social services, energy and utilities, state and local digital services, and responsible AI: work that helps the people public systems routinely fail.
- No weapons, warfare, policing, mass surveillance, or technology that profits from incarceration.
- No AI that decides whether a person gets a job, a benefit, a service, or an opportunity. It can support a human decision, as long as that human can inspect the evidence, correct what is wrong, and make the final call.
- Accessibility, privacy, security, operability, and multilingual delivery are engineering requirements, not a later phase.
CiviForm, and the repositories above. I am not consulting and I am not looking for a role. Reach me through chelseakr.com or LinkedIn.