Agentic AI

Agentic AI Consulting

Design and implementation of AI agent systems, workflow automation and integration with large language models — built to be reliable, not just impressive in a demo.

Agentic AI is powerful and easy to get wrong. The gap between a demo that works once and a system you can trust in front of real users or real data is mostly engineering: clear task boundaries, safe tool use, evaluation, and knowing when not to hand a decision to a model. I help teams design agent systems that are genuinely useful — grounded in 20+ years of shipping AI into production, not hype.

What I help teams build

• Agent design — scoping what an agent should and shouldn't do, choosing the right level of autonomy, and structuring multi-step workflows that stay controllable.
• Workflow automation — automating real business and data workflows end to end, with humans in the loop where it matters.
• LLM integration — connecting language models to your data, systems and APIs with sensible prompting, retrieval and guardrails.
• Tool use — giving agents safe, well-defined tools (including read-only access to databases and services) and validating what comes back.

A concrete example

One open project, DB Agentic Query, is a natural-language SQL agent for medical image annotation databases: it inspects the schema, generates and safely runs read-only queries, and can chart the results. It's a good illustration of the approach I favour — give the model tightly scoped, safe tools and verify every step, rather than letting it act unconstrained. You'll find it and other work in the projects section of the homepage.