Real Projects, Measured Outcomes, and AI Automations That Work

The Challenge

A Black Country manufacturer was losing an estimated two hours per shift to untracked downtime across five production lines. Legacy PLCs (mixed Siemens and Allen-Bradley) emitted data but nobody read it in real time. Root-cause analysis relied on hand-written incident logs compiled the following day.

Our Solution

We delivered a tablet-friendly Next.js shop-floor app and a .NET backend that ingests PLC data via Azure IoT Hub and an on-prem gateway. Engineering and operations leads see live OEE, downtime reasons, and an AI-generated daily summary (OpenAI) that clusters downtime events into themes. A predictive-maintenance module flags bearings and motors whose vibration patterns drift outside their baseline.

The Challenge

The group was running three separate booking systems across its clinics and fielding 300+ phone calls per day for appointments and test-result queries. Patients asked for online access; clinicians worried about compliance, identity, and out-of-hours clinical safety.

Our Solution

We delivered a unified patient portal secured with Azure AD B2C (NHS-style two-factor), integrated with clinic systems via HL7 FHIR, and shipped an Azure OpenAI-backed triage assistant configured with strict clinical-safety guardrails (no diagnosis, always routes urgent symptoms to 111/999). A product discovery phase with Liam Moses led the scope and clinical sign-off. The portal runs inside the UK Azure region with DPIA and penetration test reports on file.

The Challenge

A 28-person SaaS SME was drowning in operational overhead: the head of product spent Monday mornings writing exec reports, the support lead wrote release notes from raw PRs, and customer feedback from Intercom, Zendesk and app-store reviews went unread for weeks.

Our Solution

We shipped four AI automations: (1) a Monday executive summary bot that reads Linear, GitHub and Mixpanel and writes the weekly update in the company tone of voice, (2) a release-note generator that turns merged PRs into customer-friendly changelogs, (3) a feedback-clustering pipeline that groups 500+ tickets and reviews per week into themes with suggested actions, and (4) a slip-risk classifier that flags sprint stories likely to miss. All automations are observable — every prompt/response is logged and reviewable in a private dashboard.

The Challenge

A mid-sized UK law firm wanted generative-AI productivity for its fee earners but could not send matter data to public AI services. Their iManage document store contained 40,000+ client-confidential documents with strict matter-level access controls.

Our Solution

We deployed a fully private RAG system inside the firm's own Azure UK-South tenant. Documents are ingested from iManage with access-control metadata preserved end to end; embeddings are computed locally and stored in pgvector. Answers come from Claude 3 Sonnet on Amazon Bedrock (EU region, zero-retention). Every answer is cited, and fee earners only see content they already have permission to view.

The Challenge

A Mid Wales agricultural cooperative representing 140 family farms needed a livestock movement and traceability app that worked in areas with limited connectivity, supported Welsh language, and integrated with UK food-traceability requirements.

Our Solution

We built an offline-first React Native app using WatermelonDB for local persistence with a deterministic sync layer. Tagging, movements, medication records and mortality reports all work fully offline and sync when a signal returns. The app ships in English and Cymraeg, with data export in the format required by UK cattle/sheep movement services. Backed by a .NET API in AWS London.