Agents that do useful work
30+I design specialised agents and reusable skills that add dependable roles, knowledge and workflows rather than relying on one generic conversation.
AI & digital
I learn new technology by building with it.
Alongside adoption leadership at work, I run a personal AI environment that explores the full journey from an idea to a working system—models, agents, skills, workflow automation, integrations and training.
18MCP servers
769connected tools
30+specialised agents
80+reusable skills
01 / Connected systems
Useful AI is part of a system, not a magic box.
My lab connects models to real applications, services and data. It turns a chat interface into a practical environment for reasoning, orchestration and action.
01PersonIntent and judgement
→02AI / agentReasoning and action
→03MCPSafe connections
→ApplicationsToolsDataHome
End-to-end landscape
From hosted models to local inference.
I work across commercial and open model families, including local models on an NVIDIA RTX 3060 and advanced models through OpenRouter. That gives me direct experience of the trade-offs between capability, speed, privacy, cost and control.
- ChatGPT & Codex
- Claude & Claude Code
- Gemini
- DeepSeek
- Grok
- Llama & local models
- Advanced models via OpenRouter
Automation and orchestration
n8nI build workflow automations that coordinate models, services and events—making repeatable processes visible, controllable and easier to improve.
Custom MCP integration
769Eighteen MCP servers expose hundreds of tools, connecting agents to applications, services, data and home automation through a consistent interface.
Understanding models from inside
TrainI have trained and adapted my own models on top of foundation models, learning more about data preparation, training behaviour and model limitations.
Things I have built
The learning becomes real in working products.
These are personal projects, built to solve genuine needs and to understand the full path from product intent and data to agents, automation, infrastructure and reliable operation.
Personal finance platform
Pennies
A personal finance tool for capturing and monitoring expenditure, asset values, investments, funds and pension contributions—with long-term retirement modelling and inheritance-tax simulation.
- Financial modelling
- Portfolio tracking
- Scenario planning
Family life automation
AI Assistant
A connected assistant that helps my family interact with our smart home, find documents, automate routine tasks and reduce the burden of managing email and everyday administration.
- Smart home
- Document search
- Workflow automation
Model and workload orchestration
AI Service Controller
A control layer for my AI ecosystem that balances family requests between local models on the RTX 3060, advanced models through OpenRouter and, when justified, dedicated hosted compute for complex tasks.
- Workload routing
- Cost and capability
- Local and hosted AI
AI development optimisation
I use Graphify to make complex code and knowledge easier for agents to navigate, Playwright to test real browser behaviour, and reusable skills, agents and automated checks to improve the speed and reliability of development.
Adoption at scale
400+People moved from curiosity to practical use.
I have delivered hands-on AI learning to associates across R&D and Mars Global Services, while using my personal lab to build the practical fluency needed to ask better questions and recognise realistic opportunities.
A clear line. This personal learning and experimentation environment is not a Mars enterprise platform. Professional adoption experience and personal technical exploration are identified separately throughout.