If your problem is worth doing well, we should talk.

I design, build, and integrate bespoke AI systems — agentic, document-heavy, and operationally complex — that earn their place in production, not just the demo.

Agentic & RAG Document & Compliance AI Complex Operations Strategy & Integration
The Method

Four phases. Each one is a decision point, not a checkpoint.

01
Diagnosis
Two weeks inside your operation — interviews with the people doing the work, a clear-eyed view of your data, and a written diagnostic you own whether or not we proceed.
02
Architecture
Before a line of code, we publish the architecture: model choices, retrieval design, data flows, and the failure modes we design against. Approved before we build.
03
Build
A small, senior team — no juniors, no offshoring. Weekly working demos against measurable acceptance criteria. Your people stay in the loop the entire way.
04
Handover
Production deployment, complete documentation, and a transfer I stake my reputation on. The work isn’t done until you don’t need me.
Selected Work

A few engagements I’m permitted to describe.

Flagship
Document & Compliance AI
Days → Minutes
per conformity check

An AI system for regulated workplace-safety compliance that reads, cross-checks, and flags gaps across complex safety documentation — turning a manual, error-prone review into a fast, fully auditable one.

Agentic & Retrieval
RAG
answers from your data

Agentic retrieval systems and LLM knowledge bases that answer from your own documents with citations — including an adaptive AI tutor that responds to how each learner actually progresses.

Complex Operations
Multi-agent
coordination across systems

A multi-agent control architecture that coordinates perception, planning, and action across several independent systems working toward a shared objective.

The Principal

This is a deliberately small practice. You work directly with the person who designs and builds your system — not an account manager, not a rotating bench. When an engagement needs specialised hands, I bring in trusted collaborators I’ve worked with for years. That is what “we” means here.

Begin a conversation

The first conversation is private and unbilled.

A Principle

“The hardest part of enterprise AI is not the model. It is the quiet, careful work of making something a hundred people will trust on a Tuesday morning.”

— From my working notes
Begin

Bring me the problem worth doing well.

I take on a small number of engagements each quarter. The first conversation is private, unbilled, and rarely shorter than an hour.