Core practice
AI products, built by a team that ships its own
Six products carry our engineering today, across construction, enterprise chat, search tooling, creative generation and vision analytics. We build for other companies the way we build for ourselves.
AI Product Development
The situation
Before we start
A working model is a fifth of a product
Tenancy, permissions, quotas, billing, admin tooling and support surfaces are the rest, and they decide whether it can be sold.
Usage costs surface at renewal
Per-customer consumption is easy to ignore until it is not. We meter per tenant from the first release, so pricing is set against real numbers.
Nobody can tell if quality slipped
Model and prompt changes alter output silently. An evaluation suite ships with the product and gates every release.
What working means here
Signed first. Measured after.
The value of each one is set with you during scoping, from your data and your line. What does not change is that they are written down first and published against afterwards, pass or fail.
- 01
Evaluation pass rate per release
Every model or prompt change clears the suite before it ships, or it does not ship
- 02
Cost per tenant
Metered from the first release so pricing is set against real consumption
- 03
p95 latency
At the interface your customers actually hit, under load
How it is assembled
Stage by stage
Each one ends in something you keep.
- Step 01
Define the first release
What ships, what does not, and which customer it is for. Agreed in writing before any build.
- Step 02
Build the spine
Data model, tenancy and access control first, because those are painful to change once customers exist.
- Step 03
Add the model layer
The AI capability behind an interface, with evaluations gating every change to it.
- Step 04
Commercial surfaces
Plans, limits, billing and the admin tooling needed to run it day to day.
- Step 05
Launch and iterate
Ship to first customers with monitoring in place, then work feedback in fortnightly cycles.
Scope
What the work includes
- Product definition: the job to be done, the buyer, the first release boundary
- Application build: front end, API, data model, background processing
- Model layer behind an interface, with an evaluation suite and a rollback path
- Multi-tenancy: isolation, roles, quotas, audit logging
- Billing, plan limits and per-tenant usage reporting
- Admin and support tooling your team can run without engineering
- Launch infrastructure with monitoring and alerting
- Onboarding flows and product documentation
Deliverables
What you keep
- 01Product definition and release boundary
- 02Deployed multi-tenant application
- 03Evaluation suite for the model layer
- 04Billing and usage reporting
- 05Admin and support console
- 06Runbooks and IP assignment
Typical stack
Next.js · TypeScript · PostgreSQL · Drizzle · Redis · Stripe · Vercel / AWS · Playwright
Construction SaaS
Construction execution platform, built end to end
BuildUNIX runs construction projects for PMC firms: work moves through phase gates, site records are write-once, and snags are tracked from first report to handover. We built the platform end to end.
Read the engagement- Full stack
- application built end to end
- Write-once
- tamper-evident site records
- Field-first
- mobile flows for site teams
Start
How AI Product Development begins
Production builds $75K–180K · Embedded AI Team from $28K/month
Where an engagement lands inside its band is set by data readiness, integration depth, the reliability bar, evaluation burden, deployment constraints and usage economics. All six, explained.
Thirty minutes on the problem and where it happens, then a fixed-price pilot if it looks like a fit.
Get a costed planFAQ
Asked before signing
BuildUNIX, VectraGPT, XeoRank, AdFargo, Rankgent and Visalytix are all live products carrying our engineering. Each is linked from the products section.
No. We work on fees so the incentives stay simple and you keep your cap table.
Yes, and it is usually the faster route to revenue. We begin with a two-week assessment of the existing codebase before proposing anything.
Named engineers working inside your process for three months or more, with an itemised delivery layer rather than a blended rate. From $28,000 a month.
Often paired with
Where this fits with the rest
- 01Core AI
End-to-End AI Development
First workshop to an operated system, one accountable team. Agents, retrieval, custom models.
- 02Start here
AI Pilot
Four weeks, fixed price, signed acceptance criteria. Working software, or a written case for stopping.
- 03Operate
Managed Edge & Model Fleet
Drift, retraining, OTA rollout, device health and uptime reporting, under an agreed SLA.