AI & Innovation
Useful AI, with people in charge.
I treat AI as an engineering discipline, not a demo. These are the areas I work in, and the rules I apply to all of them.
Rule 1
The model decides what
It interprets the request, chooses the tools and drafts the work.
Rule 2
Code computes numbers
Schedules, scores and sums come from tested, deterministic code.
Rule 3
People approve risk
Anything with real consequences waits for a human decision.
Enterprise AI
AI in a company needs the same things as any other system: an owner, access control, backups and a clear answer to where the data goes.
Local LLMs
Some data should not leave the building. I run local models, connect them to a self-hosted workspace, and design assistants so that an on-premises model can replace a hosted one through configuration.
AI agents
Agents are useful when their freedom is limited. I let them collect, plan and draft, then hand the result to deterministic code, policies and people.
RAG
Retrieval only earns trust when an answer can show where it came from, so I design knowledge systems that back up what they say.
Business automation
I look for repetitive steps with clear inputs, then keep a person at the point where judgment matters. Often plain code is the right answer, and AI is only needed for some steps.
AI-assisted software development
I build with AI coding agents every day. They speed up development, testing and documentation, while architecture, review and decisions stay with me.
AI projects
Currently exploring
- On-premises models that are strong enough for production assistants
- Batch processing for repeated document tasks
- Measuring agent quality before giving agents more responsibility