Products businesses run on. Every day.

HelpMeRad — AI radiology reporting

6 WEEKS TO PRODUCTION · DAILY CLINICAL USE · ZERO ENGINEERS HIRED · CLIENT BECAME INVESTOR

A practicing radiologist prototyped report generation in a chat window and hit the wall every founder hits: a demo is not a product. We took it the rest of the way — findings-to-report generation in his voice, exam-type classification, integration with his existing PACS workflow, and a second dictation product ("eyes and ears") that lets him speak findings and get a finished, filed report.

Two products, both in daily clinical use. The engagement started as a small first milestone and has expanded phase over phase — the client is now also an investor in the company. In the client's own words: "I could get it to generate reports in a chat window, but I hit a wall turning that into something real." That wall is the practice.

HealCircle — wellness practitioner marketplace

5 WEEKS TO LAUNCH · SURVIVED LIVE-EVENT TRAFFIC · 686 POSTS MIGRATED · ZERO PHI STORED

A founder came in unsure whether what she wanted to build was technically possible at all. Five weeks later it was in production — and it held up under a live event's traffic. This is the speed the method is built for: frame the problem, validate the riskiest assumption, build under governance, ship through the deploy gate.

AI-powered CRM for travel advisors

LIVE · PAYING ADVISORS · 7 → 1,200-USER EXPANSION PATH

A founder had a working prototype and a real expansion opportunity — and the gap between them was everything unglamorous: supplier matching accuracy, onboarding, pricing architecture, observability, and the operational hardening to support a path from a handful of users to over a thousand.

We operate as the fractional product and technology leadership: shipping the product forward weekly while building the audit, monitoring, and continuity infrastructure the expansion demands.

AI-native leadership simulation — edtech partner

IN PRODUCTION · 8 AI-COACHED MODULES · 50/50 EQUITY MODEL

An adaptive coaching agent runs learners through eight modules of real-world scenarios — not courseware, not videos. Decisions have consequences, the agent remembers everything across modules, and every session produces a downloadable deliverable. The model is the interesting part: a 50/50 equity partnership — co-building, not billing. Hourly billing breaks down when the product is undefined; shared ownership aligns incentives, with the same governance shared as co-owners instead of delivered as reporting.

Fully automated solicitation triage

An M&A advisory firm's clients drown in acquisition solicitations. We built an invisible platform that reads every connected inbox in real time and classifies, labels, files, and — with six guardrails — politely declines on the owner's behalf. A solicitation is read, classified, and acted on in about one second of landing. The guardrails, one-click reversal, and audit trail were launch conditions, not add-ons — autonomy is earned before it acts, never after.

The build is multi-tenant from the foundation up, with structurally enforced client data isolation, a full audit log, one-click reversal of any action, and Google and Microsoft security verification managed end-to-end on the client's behalf. Milestones have shipped ahead of schedule, including a live full round-trip validated against a real external mail provider.

AI underwriting workflow for livestock insurance

A specialty livestock insurance agency processed every application by hand. We delivered an AI parser that reads applications and carrier PDFs, plus a rules engine encoding the carrier's real bind-authority rules — twelve of them, from per-animal dollar limits to age cutoffs to auto-refer conditions — with every flag explaining itself and citing its source rule. The rules stayed client-editable by design — the standing rule behind it: the domain’s judgment belongs in the domain-holder’s hands.

The part the client loves: the rules are his to tune. A live editing screen lets him toggle rules, adjust thresholds, and rewrite messages — no code, no waiting on us, every change audit-logged. Phase 3 shipped two weeks ahead of its date, and when the client sent a ten-item backlog, five items — including both high-priority ones — were fixed and live at $0 as warranty on delivered work.

Vet billing system — audit, verdict, hardening

A veterinary practice ran two systems: a large breeding-management platform handling $3M a month in billing, and a fast-built billing system handling real payments and DEA-logged controlled medications. A competing team was pitching a full rebuild of the big platform — and we bid on that work too. Our verdict was "don't rebuild," delivered knowing it argued against our own larger engagement. The client stayed with her rebuild team on the big platform. Then she handed us the billing system, on the strength of how we'd reached that verdict.

We treated the billing system the same way — cloned it, installed it, ran all 1,444 of its automated tests, and traced every path a dollar takes from invoice to processor to books. The verdict here: keep it, harden it. The form that verdict takes is a rule of ours now: validate the diagnosis, contest the cure, keep the standard. The core was genuinely well-built; what it lacked was the safety net — proven backups, double-charge protection on seven payment paths, a tamper-proof compliance log. We delivered a graded scorecard, a priced three-phase plan the client could stop after any phase, and then shipped Phase 0: every identified exposure path closed and verified live on production, payment records reconciled to the penny — $0 missing — and a database restore proven with a documented runbook. Then we re-assessed our own work independently and published the client her new grade.

Health inference platform — fractional CPTO

A health company with a principal engineer and a hard product needed executive product and technology leadership, not more hands on keyboards. Discovery, compliance readiness, and an engineering roadmap the team executes against with confidence — SOC 2 designed in from day one, because the evidence infrastructure was already there.

Claude Code with perfect context drifted on 4 critical architectural decisions.

We asked it to draft a product spec on our own platform, with everything an engineer could provide: full prior specs, a 200-line memory file, the active objective, every governance tool. It produced a clean, well-structured spec. The automated persona review flagged 25 findings in about 30 seconds — 4 of them critical drifts: a duplicated audit stream, a deprecated identity field contradicting a recent schema rewrite, a circular dependency on unshipped work, a compliance review at the wrong phase. Each would have been caught eventually — in code review, QA, or a production incident months later. "Eventually" compounds non-linearly.

A senior engineer approved a pull request that shipped 9 routes with no authentication.

The PR added portal pages rendering billing and contractor-rate data. It went through ordinary human review; the reviewer checked the "important" routes and default-passed the rest — the completely normal shortcut every team relies on under load. None of the nine new routes had auth middleware. The fix was not another human reviewer. The fix was a structural rule that now runs on every future PR, whether anyone is paying attention or not.

Security as a standing discipline, not a panic response.

A platform that claims to govern AI-assisted engineering should govern its own. Our git history shows nine-plus explicit security-fix commits over six months, 57 access-control tests added in a single pass, eleven route files hardened in another — none of it incident-driven scramble. Most security pages cite generic compliance language; this one cites grep-able commits in a real repository.