Skip to content

Five systems in production.

Four of these were built inside a state health service by the engineer who founded dohoff. One was a client engagement. All five are in production, and every figure below is measured rather than estimated. This work runs in regulated environments, so we name outcomes rather than organisations.

Ambulance record matching

A state health service / Enterprise, statewide

Built in-house

Delivered as AI Integration

6 months to 480ms

The problem

Ambulance arrivals and hospital records lived in two systems that did not talk. Reconciling them was manual and ran roughly six months behind, which made it impossible to see ramping as it happened.

What we built

A machine learning matching system on a concurrent processing pipeline, built in three weeks. Distributed across a multi-server cluster so coordination never stops, with event-driven ingestion and full distributed tracing.

The result

12,000 records a day, 97% matched automatically and end to end in 480 milliseconds, with no false match ever recorded. The rest go to a person rather than a guess. The six-month lag became real time, and the data now feeds a public transparency service.

Enterprise AI assistant

A state health service / Enterprise, organisation-wide

Built in-house

Delivered as AI Integration

Answers you can check

The problem

Staff could not find what they needed across an enormous internal document estate, and no general-purpose AI tool could be trusted with it because nothing could be verified or kept inside team boundaries.

What we built

An assistant grounded in the organisation own documents, with citations that jump to and highlight the exact source passage. Team-scoped access enforced so nothing leaks across boundaries, agentic routing to the right colleague, human approval before anything is sent, and results weighted to the facility the person works at.

The result

Answers staff can verify against the source, drafted emails and briefs produced to internal specification, and the whole thing running natively where people already work.

Clinical document digitisation

A state health service / Enterprise

Built in-house

Delivered as AI Integration

22 staff to 1

The problem

Handwritten medical forms were being transcribed by hand. The backlog was permanent and it took twenty-two people to hold it steady.

What we built

A system that reads handwritten clinical forms and turns them into structured records, delivered as a working solution in two days.

The result

The backlog cleared overnight. The work that occupied twenty-two people now needs one.

Workforce management

A state health service / Enterprise

Built in-house

Delivered as Web Applications

Configured, not rebuilt

The problem

Workforce systems normally force an organisation to change its processes to match the software. This one could not, because the compliance requirements were not negotiable.

What we built

A workforce platform where fields, interfaces and org structures are configured by users rather than developers. AI-assisted forms parse resumes, populate requisitions and suggest content in free-text fields. Executive dashboards embedded for real-time analytics.

The result

A system adaptable to any structure without process change, and strong feedback from HR on centralisation and day-to-day usability.

Compliance training and audit

A hospitality group / Commercial, multi-tenant

Client engagement

Delivered as Web Applications

Every record, tamper-evident

The problem

Venues have to prove staff were trained on current regulations. Generic training platforms do not know your rules, and a completion record nobody can verify is worth very little at audit.

What we built

A multi-tenant platform that takes an operator own regulations, as text, links or uploaded documents, and generates training modules with assessment questions from them. Completion produces a cryptographically hashed record the database physically refuses to alter or delete. Escalating reminders by SMS and email, evidence reports on demand.

The result

Training that matches the rules an operator actually works under, and an audit trail that holds up because it cannot be edited after the fact.

Why there are no screenshots here.

Most of this work runs inside health services, on systems that hold clinical and personal data. Those systems belong to the organisations that run them and are not ours to publish, and the same discretion will apply to your work. What we can do is walk you through the architecture, the trade-offs and the things that went wrong, in detail, on a call.

Book a call