Small clinical teams, working AI, no lock-in
Cost-lever audits, Claude Code project hygiene, single-agent stacks scoped for one-to-five operators.
We have shipped this for our own operation; we know exactly which corners are cheap and which corners bite.
Read the three cards, read the figure-free scope-bracketing, read who we say no to.
What this page is
Audience: lean operators who already ship
You are a small AI-product company, a clinical-tech startup, a founder-operator running on Claude Code. Your bill is climbing faster than you planned for. You suspect the architecture is wrong but you do not know which corner is biting. This is the work we do best.
Where this sits
Not blue-chip; not manufacturer-grade
Not blue-chip clinical-affairs work; that lives on its own page. Not manufacturer-grade regulatory consulting; that lives on Services at the Bullet Proof tier. This page is the lean-operator surface, scoped to teams who ship already and need the wiring fixed.
Scope
Three shapes the engagement takes
Cost-lever audit
Walk the actual Anthropic receipts. Tag every charge by what it really was. Identify the wiring gap, not the model-choice gap. In our experience, the three-hour fix beats the two-week rebuild in most cases we have walked through; that is a heuristic, not a guarantee, and the work is never complete in the sense of one-and-done. Worked example: our own £1,940 month.
Project-hygiene bootstrap
The database-style content architecture, the RAG mirror, the session-discipline patterns. One-week skin-in-the-game build on your codebase. You leave with the wiring, not the slide deck.
Single agent stack
One bespoke agent, deployable behind your own keys, your own data, your own ethics gate. The shape we use for the GNL Grace stack itself. Built to be deletable on day one, not to lock you in.
Tier
Light Touch or Strong Signal, scope first
Light Touch (one workstream, one deliverable, 20 to 40 hours) or Strong Signal (working artefact at the end, 80 to 160 hours). Full tier description on Services. We cost from scope, not from a price sheet; the case study walks the receipts on our own work so you can see the shape of the engagement before any conversation about figures.
Where the substrate is right, we partner long-arc on lactate as the canary in the coal mine for metabolic disease, on bolus advisors that need IOB modelled honestly, and on AID systems adding exercise or IOB features in lockstep with the evidence. We carry the same posture in lean engagements.
The refusals
Who we say no to
Teams whose AI bill is genuinely small. The wiring discipline is overkill, and we will tell you so on the first call.
Pre-product teams. We work with operators who already ship; not with teams still building MVP zero.
Teams hoping for a flat 80% bill drop on a glossy slide. The honest number depends on team size, session intensity, and what fraction of developer time is read-class versus build-class.
Talk to us
One inbox, scope first, figure follows
Read the case study before the call if you want to see the shape of the work. The first conversation is about whether the engagement fits the substrate.
Start a brief
Read the case study first
One inbox: john@theglucoseneverlies.com