I build AI agents which do useful things.

## what I build

• agentic AI Agents that monitor your systems, make decisions, and execute tasks — with you in the loop. → replaces: manual monitoring, copy-paste workflows, being the human middleware between your systems
• data readiness Your data is in 6 spreadsheets, 2 databases, and someone's email. I clean it up and wire it into a fast layer that agents can actually use. → replaces: data scattered across systems, manual reports, "I'll have those numbers for you by Friday"
• connectors Your team already lives in Claude or ChatGPT. I wire your real systems — docs, metrics, CRM, internal tools — into those chats via MCP, so AI can read, search, and act, not just guess. → replaces: copy-pasting context into chat, "the AI doesn't know our business," paying for subscriptions that can't see your data
• AI triage You tell me what hurts, I tell you what an agent can fix, what it can't, and how. You get a build-plan, not a slide deck. → replaces: "we should use AI for something," trying to figure out where to start
• custom apps Software that handles the routine and flags what needs you. Built around how you actually operate. → replaces: off-the-shelf tools that almost fit, alt-tabbing across systems, critical processes living in spreadsheets

I don't build: legacy BI dashboards, generic chatbot wrappers, or AI-as-buzzword slide decks.

This isn't a replacement for your AI subscriptions. Keep those. I build the stuff they haven't aimed their death star at yet.

## is this you?

• › Your team's context lives across 5+ tools and no one has the full picture
• › You need a working prototype to get buy-in, not a strategy deck
• › You vibe-coded something useful and need to productionize it: tests, evals, deploys, and plumbing into your Claude.ai chat and the rest of your stack

## how I work

One person holds the whole project in their head, end to end: the data, the prompts, the code, the deploy. That's what keeps agentic work from turning into AI slop.

1. Discovery call — scope the problem, go/no-go on the spot
2. Paid prototype — working demo on real data in a week
3. Production build — ship in 2-4 weeks, with weekly demos

Walk away at any phase. No lock-in. Based in Sydney.

Stop being the middleman between your systems. get in touch →

## this site is the demo

I follow the firehose of AI releases so you don't have to, and try the useful ones here first.

The chat above runs a small, cheap model at about a tenth of a cent per reply. Before any message reaches it, Jev, a decision model built for fast yes/no calls, screens out bad-faith prompts. Right model for each job, not the biggest model for everything.

## not ready to build? I do AI audits and consults too.

LinkedIn is full of snake-oil salespeople who've found the alchemical secret to turn your workflows into gold-plated automations. Meanwhile, real people are getting tired of wading through word upon endless word of LLM-generated slop.

I look at your actual stack and tell you where an AI agent saves real money, where a simple script does the job, and where the answer is a better process.