§ Case study · Bespoke build
AI-native MVP for a fintech founder team.
A working product in the market inside twelve weeks — three engineers, one designer, and a founder who could keep shipping after we left.
a fintech founder team
Client
Founder & SMB
Tier
12 weeks
Duration
2026
Delivered
The context
What they came to us with.
Two technical founders and a first hire, an idea sharp enough to draw pre-seed money, and no engineering capacity to build the first version. They'd been quoted a build by a body-shop consultancy that would have used their entire runway on the pilot.
The product was AI-first — LLM orchestration at the centre of the user experience — but they were clear that they didn't want to hire around it before validating that the shape worked with real users.
They needed a working product in the market inside a quarter, code they owned end to end, and the ability for their two engineers to keep shipping after the engagement ended.
The approach
What we did, in order.
-
Small team, senior throughout.
Two of our senior engineers alongside their two, one designer shared with another engagement. No juniors under seniors. The lead was on the code every day.
-
Written-down scope, weekly demo.
A four-page scope document produced in the first week. A working demo every Friday for both founders. Scope changes discussed in the room; three shipped, two deliberately deferred.
-
Rails backbone, TypeScript surfaces, Python for the ML side.
Ruby on Rails and Hotwire for the product surface — Turbo, Stimulus, no SPA. TypeScript for the browser-heavy flows. Python where the ML libraries lived. One deploy pipeline, one on-call rota, no microservice archaeology.
-
LLM orchestration in Ruby.
The agentic layer sat in the Rails app. Structured outputs, guardrails, eval harness, cost caps. Ownership stayed with the product team, not siloed in a Python service they wouldn't touch.
-
Handover as a phase.
From week seven, we paired more than we drove. Week eleven was code-review reversed — their engineers led the merges; ours commented. Week twelve was on-call together, then off.
How we measured
The bar we held ourselves to.
- Weekly demo against the scope document. Every deferred item logged with a reason.
- Test coverage on core product surfaces before feature launch — not a target, a gate.
- LLM output eval on a small fixture set drawn from beta-user interactions, refreshed each sprint.
- Cost-per-user projection updated weekly against actual usage; the pricing model was tuned mid-engagement based on it.
The outcome
Where it landed.
12 wks
Concept to live product
4
Engineers on the engagement (2 ours, 2 theirs)
£0
Follow-on retainer
Live in twelve weeks with paying beta users on day one. The handover completed on schedule; the client team has shipped every subsequent release themselves. The follow-on retainer was £0 — by mutual choice, not by accident.
The founders have since raised their seed round on the shipped product. The pricing model — which had been a leap of faith at week two — held up at week twenty. We stay in touch as engineers who like the founders, not as a vendor.
Retrospective
What we'd do differently.
- We over-invested in one feature the founders had asked for in week one and under-shipped a supporting flow they later called "the actual moat". We'd push back harder on scope early next time.
- The eval fixture set was drawn late — we would draw it from week two, from paper prototypes if necessary.
- We would insist on the designer being full-time from week one. Sharing them across engagements slowed us in the first three weeks.
Services used
The moves in this engagement.
S1 · Build
AI-Aided Development
Senior engineers paired with AI coding agents, custom tooling, and evaluation harnesses — applied to your delivery or operationalised in your team.
S3 · Build
LLM Integration
Chat, RAG, and the wider integration stack — measured, regression-tested, and cost-aware from the first prompt.
S8 · Build
Bespoke Software Development
Full production software delivery, usually with AI at the core — fixed team, named engagement lead, no staffing pyramid.
Related work
Other engagements in the same shape.
§ Case study · Agentic systems
An agentic ops-triage copilot inside a UK retail bank.
Analyst-in-the-loop triage across three legacy ticketing systems — mean time to first response cut by more than half, without a single new ticketing surface.
§ Case study · LLM integration
Editorial-assist RAG for a global media company.
A retrieval-assisted drafting surface for a 300-strong editorial team — answers cited to the archive, hallucinations effectively eliminated on tested queries.
Have something with this shape?
A 30-minute discovery call. We'll say honestly whether the pattern fits.