AI Consulting & Implementation

AI that survives
contact with production

I build systems that hold up against real users, real data, and real edge cases. Not prototypes that only impress in the demo.

Start a conversation
An honest no when AI isn't the answer
Systems that ship, not slide decks
Clear pricing before you commit

Built AI & software for

As consultant and engineer, before and through Visengence

  • Synthpop Healthcare AI
  • Ingemark
  • 3MI Lab
  • ClickOut Media
What I do

Three ways I can help

Strategy when you need direction. Implementation when you need to ship. Data engineering when your foundation is shaky.

01
01

AI Strategy

Figure out where AI actually makes sense for your business, and where it doesn't. No hype, just honest analysis.

Learn more
02
02

AI Agents & Automation

Multi-agent systems that handle real workflows, not toys that break when users do something unexpected.

Learn more
03
03

Data & ML Engineering

Turn messy data into models that work, and models into systems that keep working six months later.

Learn more
What I've learned

Why AI projects fail

01

Wrong problem

Building AI for tasks that don't need AI. Solving symptoms instead of root causes.

02

Data isn't ready

Messy, siloed, or insufficient data. The model is only as good as what you feed it.

03

No one owns it

IT thinks it's a business problem. Business thinks it's an IT problem. Nothing ships.

04

Demo ≠ Production

The prototype works. Then edge cases, scale, and real users break everything.

I've made some of these mistakes myself. That's why I built a process to catch them early.

Selected work

Built, shipped, running

Recent enterprise work, anonymized. I'm happy to talk specifics on a call.

01 · Enterprise IT department

AI assistant for level-1 IT support

A support team buried in the same routine requests every day: password resets, access requests, known errors. Skilled people doing work a runbook could describe.

What shipped

An AI assistant that handles level-1 support end to end: it resolves routine requests itself, asks clarifying questions like a human agent would, and escalates cleanly to people when it should not decide alone.

What changed

Routine requests get resolved without a human in the loop. The support team's time goes to problems that actually need people.

02 · Complex enterprise integration landscape

Debugger for multi-system integrations

When something broke between systems, engineers spent hours tracing a single error across logs, queues, and services owned by different teams.

What shipped

A debugging assistant that follows an error across system boundaries, correlates the evidence, and presents the likely root cause with the trail that supports it.

What changed

Tracing one cross-system error dropped from a half-day of manual log-digging to a short review of a pre-assembled case.

03 · Company-wide internal platform

One AI layer over all enterprise tools

Employees jumped between five different apps to answer one question, each with its own login, search, and quirks.

What shipped

A unified AI layer that lets employees talk to all their enterprise tools from one place, with permissions respected per user and per system.

What changed

One conversation replaces five app switches, with each employee seeing only what their permissions allow.

How I work

No magic, just method

I've seen enough AI projects fail to know what actually works. Here's how I de-risk yours.

1

Understand the mess

I dig into your actual workflows, data, and pain points. Not a questionnaire: real conversations with the people doing the work.

2

Find the lever

Most AI projects fail because they solve the wrong problem. I look for the one workflow where AI removes real, repeated cost, not a marginal tweak to everything.

3

Build something small

Start with a focused proof that works end-to-end. Real data, real users, real feedback. No six-month "research phases."

4

Scale what works

Once it proves out, I harden it for production. Monitoring, error handling, the boring stuff that keeps systems running at 3am.

Want a straight answer about AI?

Tell me what you're trying to solve. I'll tell you whether AI can help, what it would take, and what it would cost. If it won't help, I'll say so.

Start a conversation