AI Technical Program Portfolio

Dev Instincts

Dev Instincts turns every defect found in AI-written code into a check that can be reused. It holds 63 approved instincts, ranked by real cost from a ledger of 627 findings across 3 projects (my private procurement platform, built with AI; Ferrite, an open-source editor written by AI; and Django, as a human-written reference) and 57 review rounds.

Passing test suites and AI reviewers kept missing the bugs that cost the most: double submits, races, partial saves and stale screens. Each one got fixed once and then showed up again somewhere else.

I own the program and approve every proposed instinct before it joins the standard.

How it works

1. Every finding from every project goes into one ledger with its severity, area and the instincts that should have caught it.
2. A stats tool ranks instincts by cost. Data loss and security weigh 8, cosmetic issues weigh 1, and regressions count double. Changes get checked in that order.
3. New instincts start as proposals with evidence. I approve them before they get a number.
4. Three agents work separately. An architect agent writes the plan and the acceptance tests with expected values worked by hand. A builder session writes the code. A reviewer agent that didn't write it checks it.

Decisions

• Separate roles instead of more agents. I turned down a manager, lead and junior chain. Writing code isn't the slow part. Checking it is, and every handoff drops detail.
• Outside data. I mined Ferrite, an open-source editor of 133,000 lines written entirely by AI. It gave 154 bugs from that project's own records, 14 hidden bugs nobody had recorded, and 9 new instincts. I filed 6 of those bugs with the project, each with code lines and a probe.
• Graded sources. I collected 35 published sources on code written by AI and graded each from strong to anecdotal. A check backed only by weak sources stays a proposal until the ledger supports it.
• Stop rule. If defects don't roughly halve each pass, or regressions go above about 10% of fixes, patching stops and the code gets restructured or rebuilt from the baseline.

Results

Documents produced: the instinct standard, findings ledger and stats, convergence tracker, audit spec, review and verify templates, agent definitions, two case studies.

Full project page

CertMoat and Tahdoo

As founder of Inextier LLC (Jun 2025 to Jul 2026), I took two desktop apps through Microsoft Store certification over 16 packaged builds. I ran them in sequence so the second app got every fix from the first.

• Tahdoo is an offline study app with notes, tasks, calendar, flashcards and optional AI tools. It has a 7-day trial tracked on the device and a Store subscription. 9 packaged releases.
• CertMoat is an air-gapped CMMC and HIPAA compliance workbench for small contractors and medical practices. Its 171 controls are stored as validated data files. It runs gap analysis, tracks evidence and generates the System Security Plan. 7 packaged releases.

I was product owner and solution designer, covering requirements, architecture, release order and certification.

Decisions

• Find the root cause. Tahdoo's early submissions failed. I traced each failure to its platform cause and wrote it up before resubmitting, and added a full review before every package.
• Release order. I held CertMoat until Tahdoo passed, then reused its licensing setup and purchase helper with separate product IDs so the licenses can't collide.
• Cut scope. I dropped a Stripe and license-server billing design for Store licensing only, then shut down the servers it needed.
• Offline by design. CertMoat blocks network calls by default, checks its own SHA-256 hash when it runs, and logs every change to a keyed-hash audit log.

Both apps passed Microsoft Store certification, and the CertMoat trademark application was filed with the USPTO. CertMoat is now being redeveloped for sale.

What I learned: Platform certification is a test suite you don't control. Reproducing each rejection was faster than resubmitting and hoping.

Full project page