Case studies

Working projects, honestly described.

We'd rather show you how we build than list technologies. Each project covers the problem, architecture, tools, status, limitations and value — without overstating production readiness.

Multi-Agent TestAI

Flagship project

Working prototype · Jira/Zephyr integration planned

Overview: An AI-powered QA system that generates, evaluates and improves test cases using cooperating agents.
The problem: Writing and maintaining good test cases is slow, inconsistent between people and hard to keep in step with changing requirements.
Approach & architecture: A planner agent breaks requirements into testable behaviours, a generator agent drafts test cases, and a reviewer agent scores them for coverage, clarity and duplication, sending weak cases back for refinement. A human approves before anything is exported.
Tools: PythonFastAPILLMs (local and cloud)Multi-agent orchestrationEvaluation scoring
Status & limitations: Prototype stage. Jira and Zephyr integrations are planned, not live. Output quality depends on requirement quality and still needs human review.
Business value: Faster first drafts, more consistent test documentation and an auditable review loop — the basis for the TestAI Command Center product.

Local AI Infrastructure

Engineering capability

Operational lab environment

Overview: A local-first, on-premises-ready LLM platform using Docker, Ollama, FastAPI, OpenWebUI and MCP-based integration patterns.
The problem: Many organisations want AI assistants but cannot send sensitive data to public AI services, and don't know what running models privately involves.
Approach & architecture: Containerised model serving with Ollama, an API layer in FastAPI, OpenWebUI for users, and MCP servers to connect models to tools — all reproducible with Docker.
Tools: DockerOllamaFastAPIOpenWebUIMCPRAG
Status & limitations: A lab and reference environment rather than a client production deployment. Local models trade some capability for privacy and need hardware planning.
Business value: A proven reference architecture for private AI deployments and an honest view of the cost, privacy and maintenance trade-offs.

DxLens

AI product innovation

Research prototype · not clinically validated

Overview: A smartphone-assisted diagnostic test reader explored as a research prototype.
The problem: Reading rapid diagnostic tests by eye is subjective, and results are hard to record consistently outside a lab.
Approach & architecture: Phone-camera capture with image normalisation and an AI model that interprets test-strip results and records them in a structured form.
Tools: Computer visionMobile captureImage processingPython
Status & limitations: Status: Research prototype only. Results shown are simulated and are not clinically validated. Not a medical device and not for diagnostic use.
Business value: Demonstrates end-to-end AI product thinking — from capture to interpretation to responsible limits.

Have a similar challenge?

We can show you the architecture behind any of these projects on a call.

Start a conversation Ask Jay