What you own afterwards.Deployed in your estate. Source transferred on completion.
Four kinds of system, all built to the same standard and all handed over. No per-seat meter, no platform lock-in, no renewal that doubles. If you ended the relationship tomorrow, the thing we built would keep running and your team could maintain it.
Four kinds of system.
Each of these is a category we have shipped into production, not a capability we could theoretically offer.
Multi-step work, made replayable
Agents built as explicit state machines rather than free-running loops, with typed tool schemas, bounded retries and hard budget ceilings. Every run writes a full trace — inputs, tool calls, intermediate state, the decision point and who approved it.
- Typed tool schemas
- Explicit state machines
- Bounded retries & budgets
- Deterministic replay logs
Extraction at volume, with a floor
High-throughput document processing where the output is structured, checked and traceable to the page it came from. Below a defined confidence floor the item goes to a person rather than through.
- Extraction at volume
- Structured, validated output
- Confidence floors
- Human review queues
Answers bound to their source
Chunking tuned to the shape of your documents, hybrid sparse and dense retrieval, a reranking pass, and citation that is a property of the pipeline rather than something the model was asked to produce.
- Structure-aware chunking
- Hybrid sparse + dense retrieval
- Cross-encoder reranking
- Permission-aware indexes
Bespoke AI SaaS, under your brand
A whole software product with AI at the core — multi-tenant, authenticated, billed, deployed on your domain. Ordinary engineering done well is most of what makes an AI system usable.
- Multi-tenant architecture
- Your brand and domain
- Test coverage before handover
- Built to scale past pilot
Selected production builds.
Shipped across healthcare, retail, fintech, media, SaaS and productivity. Everyone who worked on these wrote code on them.
HIPAA-compliant transcription
Regulated audio through speaker diarisation and multi-format export, behind scoped OAuth. Storage chosen for compliant handling rather than convenience.
Deepgram · Fly.io · Supabase · Next.jsIn-store AI avatar
Camera detection driving a real-time greeting on a shop-floor screen over a WebSocket stream — containerised so the same build runs in every store.
HeyGen · Python · Docker · ReactPayment platform rebuild
Rebuilt with more than 500 tests, which surfaced around twenty billing defects in discount-stacking logic that had been live and unnoticed.
React · Supabase · Stripe Connect · VitestSecure cloud storage MVP
Authentication, nested folders, in-browser file viewing and usage-based billing — specified, built and shipped in seven days.
Next.js · GCP · Stripe · ShadcnThree-stage image pipeline
A chained model pipeline — segment, inpaint, composite — rather than a single prompt and hope. Each stage independently testable.
Stable Diffusion · Next.jsPresentation SaaS prototype
CSV-driven deck population with native .pptx export — the unglamorous format work that decides whether a tool gets used.
Next.js · ReactWhat handover actually means.
Ownership is a contractual and technical fact, not a reassurance. This is the list we work to.
- Source code transferred to your repository on completion
- Infrastructure runs in your cloud account or your data centre, under your billing
- No SimpleAI licence key, callback or phone-home in the deployed system
- Architecture decisions written down, not held in someone's head
- Test suite handed over and runnable by your team
- Runbook covering deployment, rollback and incident response
- Evaluation harness and held-out set, so quality can be measured after we leave
- A named engineer available during handover, and a defined support window after it
Common questions.
Do we really own the source?
+
Yes. Source is transferred to your repository on completion and the deployed system contains no licence check, callback or dependency on us. If you ended the relationship the day after handover, the system would keep running and your team could maintain it.
What if we want you to keep operating it?
+
That is Managed AI, on a monthly retainer — monitoring, drift detection, retraining, cost control, human review queues and incident response. Ownership and operation are separate decisions and you can change your mind about the second one at any point.
How long does a first build take?
+
Typically four to twelve weeks from signed scope to live, driven mostly by deployment model and regulatory load rather than the AI itself. A cloud-deployed document workflow sits at the fast end; an air-gapped system in a regulated environment sits at the slow end.
We can help.
Tell us what's hard, expensive, or taking too long — and we will help you identify, optimise and deploy AI for innovation, efficiency and growth.