Start with the real problem
Be clear about the decision and where AI should—and should not—be involved.
Applied AI in practice
I turn reasoning models into governed operational systems.
I define what the model may infer, what authoritative services must calculate, what evidence supports each decision and where human control remains essential.
Where I add value
How I take AI from an idea to something people can rely on.
Most of the hard work sits around the model: understanding the job, getting the facts straight, building the right tools, setting boundaries, testing and keeping it working.
Get the facts straight
Use trusted data, track where it came from and flag gaps or conflicts.
Put each job in the right place
Let software do the calculations, the model interpret and people make the important calls.
Set clear rules
Tell the model how to handle uncertainty, missing evidence and conflicting information.
Give it the right tools
Provide controlled access to data and actions instead of giving it free rein.
Test what can go wrong
Check poor data, edge cases and failures—not just the happy path.
Run it properly
Add security, monitoring, recovery and a safe way to improve it.
Keeping AI under control
Do not ask the model to do everything.
A model is good at making sense of information. It should not be the source of truth, do untested calculations or decide for itself what it is allowed to change.
The rules I expect a reasoning model to follow
- Separate facts from assumptions and recommendations.
- Check current information before reaching a conclusion.
- Use tested software services for calculations and official status.
- Say when the data is poor, incomplete or uncertain.
- Do not guess when the evidence is missing.
- Ask before making important changes or taking action.
- Show where the important facts came from.
- Stop safely when identity, data or validation checks fail.
Selected work
Examples of how I use AI in real work.
These examples show AI being used as part of the work—not as a replacement for judgement, but inside a system with trusted data, clear limits and human control.
CASE 01
Public platform
Making sense of endurance data
Montis makes the coaching decision. AI provides the narrative.
- Problem
- Training data comes from several places. The challenge is to turn it into a reliable coaching decision without handing that decision to a language model.
- What I did
- Montis builds the coaching decision from validated data and explicit rules. AI turns that result into a clear explanation and conversation. It does not invent the athlete’s training state or decide the coaching direction.
- My contribution
- I designed and built the product, training models, tools, integrations, security, validation and production services.
CASE 02
Anonymised
Making sense of a poorly documented building
- Problem
- The building system worked, but nobody had a complete, reliable view of how rooms, switches, devices and control addresses fitted together.
- What I did
- I combined live bus data with the available drawings and records. AI helped suggest likely matches, but only checked and confirmed relationships became part of the source of truth.
- Why it matters
- AI can help narrow down a messy technical problem without pretending every suggestion is a fact.
CASE 03
Anonymised
Finding the evidence in a difficult technical project
- Problem
- Important facts were spread across technical documents, emails and conflicting accounts from different people.
- What I did
- I used AI to sort the material, compare claims and find contradictions. Each important point stayed linked to its source and was marked for confidence and human review.
- Why it matters
- The result is useful decision support that can still be checked and challenged.
CASE 04
Anonymised
Using AI in everyday engineering
- Problem
- My work crosses application code, cloud services, live data and documentation. Different AI tools are good at different parts of that job.
- What I did
- I use different reasoning and coding tools to research, diagnose, implement and review. I give each one the context and access it needs, then check its work against the real system.
- Why it matters
- The value is not asking one model to do everything. It is knowing which tool to use, what to trust and where I need to make the final call.
Show the work. Protect the client.
These examples explain the problem, the choices and the outcome. Names, locations, identifiers and sensitive technical details have been removed. Where needed, example data is made up and clearly labelled.
The point is to show how I worked, not to publish private material or raw AI conversations.
Practical Applied AI
Need to make AI useful in the real world?
I can help turn an idea into a working service, or bring control and clarity to something already underway.