I'm a public-interest software engineer building trustworthy data systems for transit agencies, nonprofits, and communities, mainly in Python and TypeScript.

I work where delivery, evidence, and public accountability meet: accessible interfaces, privacy by design, reproducible analysis, fail-closed checks, and explicit limits on what a system can prove.

Based in Davis, California · Portfolio & consulting · LinkedIn

This independent R&D portfolio began in June 2026. Unless marked otherwise, projects are pre-1.0 beta or reference implementations; each repository documents its evidence, known limits, and review or deployment gates.

• Transit Delivery Atlas — Independent, source-linked crosswalk of 21 California transit directives, with named entities, scoped timing, delivery dependencies, and open questions kept separate from the signed source. Explore the live atlas.
• GTFS Scorecard — Live, plain-language quality scorecards for 1,100+ U.S. and Canadian GTFS feeds, with GTFS-Realtime monitoring where available, a fail-closed CI gate, and a read-only MCP server. Visit gtfsscorecard.org.
• tods-validate — TODS v1.0–v2.1 validator for CLI, GitHub Actions, pre-commit, browser, Docker, and LSP workflows, with auto-fix and GTFS drift diagnostics. Includes a merged upstream standards fix.
• NearMiss — Reproducible road-hazard analysis demonstrated with synthetic Davis and Riverside data, known-answer hotspot benchmarks, multi-source ingestion, and a QGIS plugin. Explore the live project.
• Fare Policy Assistant — Reduced-fare reference assistant for five California transit agencies, designed for cited EN/ES answers with a public 201-case evaluation and a separate black-box audit.
• Swelter — Replicable neighborhood heat and air-quality sensing with calibration tooling, accessible bilingual maps and alerts, clearly labeled real-data demos, and OGC SensorThings export. View the demo.

• Keep source facts, calculations, and analytical judgment visibly separate.
• Make accessibility, privacy, provenance, and correction paths part of the architecture.
• Prefer reproducible evidence and known-answer tests over persuasive demos.
• Treat uncertainty and negative findings as product requirements, not footnotes.

My broader portfolio includes nonprofit record reconciliation, reproducible funder reporting, tenant and community evidence systems, queer digital preservation, environmental monitoring, and privacy-first personal software. Browse the full portfolio or see all repositories.

I use AI agents as implementation collaborators while retaining responsibility for architecture, standards, threat models, evaluations, acceptance criteria, and release decisions. Legal, policy, subject-matter, and manual accessibility reviews remain human gates and are labeled when pending.

I'm especially interested in engineering roles and collaborations involving transit delivery, civic data infrastructure, accessible government technology, and trustworthy AI. For roles, consulting, collaboration, or a thoughtful technical conversation, reach me through chelseakr.com or LinkedIn.