AI solutions on the Microsoft stack.
Every client wants AI. Almost none of them need the same AI. My job is telling those situations apart, then building the right one.
What it is
The current chapter of my work at KPMG: demoing, researching and building AI solutions on the Microsoft ecosystem. The toolbox spans Microsoft AI Foundry for custom model-backed solutions, Copilot Studio for conversational agents, and Agent 365 for agentic workflows inside the Microsoft 365 estate.
The consulting half of the job matters as much as the engineering half: mapping a client's business problem to the right level of the stack. Some problems want a managed agent, some want a custom Foundry build, and some want no AI at all. Saying which is which, with evidence, is the real deliverable.
The work
- Building and demoing AI solutions with Microsoft AI Foundry, Copilot Studio and Agent 365
- Researching the fast-moving Microsoft AI stack and translating it into client-ready guidance
- Consulting clients on which solution actually fits their business problem, and which does not
- Delivering with the good practices absorbed along the way: git discipline, CI/CD, Agile ways of working
Concepts
What it taught me
The hardest part of enterprise AI is not the model. It is matching the tool to the problem and having the honesty to say when the answer is a simpler one.