Shipping
Jul 2A11yReadyThe self-improving loop now reaches the pipeline's prompts, not just its code. Many…Jun 30A11yReadyThe system now diagnoses and fixes its own recurring problems. The Pipeline Advisor was…Jun 29Testing PlatformSelf-serve admin-perms diagnostic landed: 4 role codes unblock 32 specs, verified…Jun 29Testing PlatformThe platform automatically surfaced 5 real legacy bugs in UCR QAT — including 3 that…Jun 29Testing PlatformHonest Monday baseline: 66.2% pass excl. skips on the cleaned 210-spec / 324-test suite…Jun 29Testing PlatformThe strict PageObject allowlist now actually constrains the live generator. The v3.2…Jun 28A11yReadyTwo reliability improvements. (1) Image sizing: images inside converted documents —…Jun 28File IntelligenceRetired the Xingwu PHP-audit cross-check — it was the initial one-time orphan-detection…Jun 28Testing PlatformEight of nine holistic test categories now populated on prod (was six Friday). Sunday…Jun 27A11yReadyLibrary redesigned as a compact, sortable list (was large cards) — far more documents…Jul 2A11yReadyThe self-improving loop now reaches the pipeline's prompts, not just its code. Many…Jun 30A11yReadyThe system now diagnoses and fixes its own recurring problems. The Pipeline Advisor was…Jun 29Testing PlatformSelf-serve admin-perms diagnostic landed: 4 role codes unblock 32 specs, verified…Jun 29Testing PlatformThe platform automatically surfaced 5 real legacy bugs in UCR QAT — including 3 that…Jun 29Testing PlatformHonest Monday baseline: 66.2% pass excl. skips on the cleaned 210-spec / 324-test suite…Jun 29Testing PlatformThe strict PageObject allowlist now actually constrains the live generator. The v3.2…Jun 28A11yReadyTwo reliability improvements. (1) Image sizing: images inside converted documents —…Jun 28File IntelligenceRetired the Xingwu PHP-audit cross-check — it was the initial one-time orphan-detection…Jun 28Testing PlatformEight of nine holistic test categories now populated on prod (was six Friday). Sunday…Jun 27A11yReadyLibrary redesigned as a compact, sortable list (was large cards) — far more documents…

The Program

5 production AI initiatives, 1 embedded architectprogram

I currently run the AI program inside Multnomah County — architecting, building, and shipping production AI across 5 simultaneous initiatives. Plus other client work shipped in the last few years.

Live ProgramShipping Weekly

Multnomah County AI Program

Four production AI initiatives running simultaneously inside the County, plus one in discovery and one restricted engagement proposal. I'm the embedded AI architect: I design the systems, write the code, capture the institutional knowledge, and ship features weekly. The program runs its own public presentation hub with a dated activity feed — anyone can verify the velocity.

2
Production Systems
1
Active Development
1
In Discovery
1
Restricted Proposal

Program Initiatives

Live hub
A11yReady
In Production · Scaling

A11yReady

AI-Powered Document Accessibility Platform

DCA

Converts public-facing PDFs into fully accessible HTML in 30–90 seconds — versus 1–4 hours of manual remediation. A Vertex-AI pipeline routes each document through one of 34 type-specific clusters with a deterministic forms path that anchors fields to embedded AcroForm widgets, so extraction is stable run-to-run.

30s
PDF → HTML
95%+
WCAG 2.1 AA
$0.50–$2
Per Document
Vertex AIGemini 2.5 Flash/ProClaude Sonnet 4.5Claude Haiku 4.5+7
View details
File Intelligence Platform
In Production · Scaling

File Intelligence Platform

Drupal-Aware Site Intelligence for multco.us

Multnomah County · WDX

Nightly Google Cloud pipeline that reconciles 56,500+ Drupal CMS files across 9,900+ published pages on multco.us. Six-bucket file classification, axe-core WCAG 2.1 AA scoring, Vertex AI service classification + remediation proposals, citizen-journey personas, per-department health trends, and an editor-attribution layer so every issue has an owner.

56.5K
Files Classified
9.9K
Pages Crawled
92.3%
Xingwu Agreement
Python 3.11FastAPIReact 19Vite+10
View details
AI-Powered Testing Platform
Active Development

AI-Powered Testing Platform

6-Agent System for Legacy Modernization

DCHS / QA

A team of 6 specialized AI agents that indexes legacy codebases, captures institutional knowledge before retiring staff walk out the door, generates regression tests from that knowledge, and watches for drift on every commit. Multi-tenant from day one — UCR is tenant 1; ACHP onboards next.

5/6
Agents Closed-Loop
9,420
Indexed Symbols
148
Specs / Run
Vertex AIClaude Sonnet 4.6PlaywrightFastAPI+8
View details
UCR Modernization
Discovery

UCR Modernization

Behavioral Test Capture from Retiring SMEs

DCHS

Vision for modernizing the Universal Client Registry — DCHS's system for managing client demographics, enrollments, and services across four divisions. Gated on first proving behavioral equivalence: every workflow Rikki documented over 36 years becomes an executable test the modernized UCR must pass before it ships.

704
SME Knowledge Sessions
208
Behavioral Tests (v2)
32 / day
Questions Answered
Playwright (regression suite)FastAPIReactVite+7
View details

Earlier Multnomah County Work

Other Production Systems