Case studies
Work you can check.
One real system, with its specification, its decision log and its current build state served by an MCP server you can connect to yourself. Then two worked examples showing the shape a small engagement takes.
In build · our own product
Truckerbook
Bookkeeping for owner-operators and small carriers: settlements in, fuel and expenses captured, and an IFTA return out the other end. Regulated output, real money, multi-tenant from the first migration — the kind of system where being approximately right is the same as being wrong.
It is not a client project, and that is exactly why it is here. On client work you get a screenshot and a testimonial. On this one the specification, the architecture decisions and the build state are all live, addressable, and answer to a tool call. You can audit the claim instead of taking it.
- What
- Bookkeeping SaaS for US trucking — settlements, fuel, IFTA
- Role
- Co-founder and CTO. Architecture, the whole build, the infrastructure
- Stack
- TanStack Start · PostgreSQL · Docker Compose · Caddy · S3-compatible storage
- Status
- Phase 2 — capture core. In build, not yet launched
- The spec is machine-readable. A 237-question register and a decision log served over MCP, so an agent working on the code reads the same source of truth a person does — and cannot quietly invent a different one.
- The definition of done is enforced by CI, not by discipline. Business logic outside a service fails a test. A new table that hasn't been classified for the admin panel keeps the build red until somebody decides.
- The founders' panel physically cannot read tenant data. Separate app, separate origin, its own auth, and a database role with zero grants on tenant tables — a bug in the panel is not a data breach.
Numbers on the right are live tool output, captured 24 Aug 2026.
Worked examples
Not clients. These are the two shapes a small build usually takes, written out end to end so you can see what one involves. The figures are illustrative.
-
Barber / salon Example
From phone-tag booking to 24/7 self-serve appointments
A two-chair barbershop replaced phone-and-notebook scheduling with an online booking system and automatic reminders.
Before Every appointment came in by phone or walk-in. The chair sat empty during missed calls, double-bookings happened on busy Saturdays, and there was no record of who was coming in.
After A booking site where clients pick a barber, service, and time 24/7. Automatic text reminders cut no-shows, and a simple dashboard shows the day at a glance. Customer list lives in one place — ready for AI assistants to answer "are you open Sunday?".
-
Electrician / trades Example
Leads, jobs, and invoices in one place instead of scattered texts
A solo electrician moved quotes, scheduling, and invoicing off sticky notes and texts into one simple job manager.
Before Leads arrived by text and got lost. Quotes lived in a notebook, scheduling was guesswork, and invoices went out late — if at all.
After A lightweight job manager: capture a lead, schedule the visit, track the job, and send the invoice — from a phone on site. Customers and history in one place.
If you want the version of this conversation that ends in a number, describe the problem and I'll tell you what it takes.
Start a conversation