Three things, end to end.
Business automation, AI integration and numbers you can trust. Same two people from the first call to the handover; same four gates every time; everything in your own accounts from day one.
AI put to work inside the business you already run
Reading, sorting, suggesting, drafting — the routine work someone re-does all day. What it must get right is agreed in writing first.
- Team
- Both founders
- A queue of orders, messages or claims that someone reads all day
- Questions customers ask a hundred times a week
- A ChatGPT experiment that should become real work
- The test it must pass, agreed in writing before the build
- A working system in your own accounts, on real traffic
- A running-cost sheet: what it costs per month at your volumes
- A handover file your team can run it from
For your technical adviser
Retrieval, agents, fine-tuning, guardrails, observability. The evals are published alongside the model, before the build starts.
- Eval set and baseline metrics, published before build
- Production system in your cloud account
- Cost model per request, reviewed with the code
- Runbooks, drift alerts, exit review
Stack Claude · GPT · Gemini · Llama / Mistral self-hosted · pgvector · Qdrant · Python · TypeScript
Orders, stock, billing and follow-ups on one screen
The retyping stops. The tools you already use start talking to each other, and the owner sees the whole day on one page.
- Team
- Both founders
- Orders arriving from more places than the team can retype
- Stock in a spreadsheet that never matches the shelf
- Software the business has outgrown
- Working software your team is trained on while we are in the room
- One page every morning: sold, shipped, returned, running low
- A written decision log and tests the next engineer can pick up
- A weekly note in plain words: built, decided, why
For your technical adviser
Web, API and data products: typed, tested, documented. Fortnightly increments and an architecture decision log your team owns after hand-over.
- Working software in fortnightly increments
- Architecture decision log in your repository
- Test suite and CI pipeline that your team owns
- Written weekly demo notes
Stack TypeScript · Python · Rust · PostgreSQL · React · AWS · Azure · GCP
Every report from one place — and it matches the godown
One truthful picture of stock, sales and money, checked against reality before anyone relies on it.
- Team
- Both founders
- Reports that disagree depending on who pulled them
- Month-end that takes a week of copying between systems
- Wanting automation or AI, but not trusting the data under it
- One place every number comes from, in your own accounts
- Proof, run in parallel, that the new numbers match the old reality
- A monthly running cost you saw in writing before the build
- The owner’s morning page: stock, sales, money
For your technical adviser
Lakehouse-first warehouses, pipelines and governance. Vendor-neutral and cost-modelled before build: lineage map, parallel-run reconciliation, FinOps alerts, decommission schedule.
- Lineage map and migration plan with costs
- Iceberg or Delta lakehouse with IAM and governance
- Parallel-run reconciliation reports
- FinOps alerts and a decommission schedule
Stack Apache Iceberg · DuckDB · dbt · Spark · PostgreSQL · Airflow · Terraform
Start with a two-week discovery sprint.
Fixed fee, both founders. You get a findings document, a prototype and a costed build plan, and you keep all three whoever you build with.
Write a paragraph. A founder replies within one working day.
A 30-minute call with both founders follows. No deck, no proposal template — a first read of whether we are the right shape for the problem.