Ⅰ
Demos aren't deployments
A chatbot on a landing page is not an AI strategy. The value lives inside your actual workflows — inboxes, quotes, orders, tickets — where hours leak every week.
Applied AI consulting · teams of 10–50
Your competitors are reading about AI.
You'll be running it.
I embed with your team, find the workflows worth automating, build the agents that automate them — and leave your people fluent, not dependent. Frontier models or local open-weight ones, picked per job, not per hype cycle.
- 0
- operating frontier models & agents daily
- 0
- Oracle Cloud ERP · OTC · P2P · integrations
- 0
- the team size I build for
- 0
- person you'll ever deal with
01
The gap isn't access to AI.
It's application.
Every business of 10–50 people has the same three problems. None of them are solved by another subscription.
Ⅱ
The tools change weekly
Claude, GPT, Kimi, GLM, Codex — the frontier moves faster than any team can evaluate while doing its day job. Picking the wrong model costs more than picking none.
Ⅲ
Your data can't just leave
Client records, financials, IP. We scope exactly what touches an API — and when truly nothing may leave the building, a local open-weight model on your hardware does the job. I'll tell you when that's overkill.
02
Work that ships.
Four ways to engage. All of them end with something running in your business — not a slide deck.
Workflow Audit
1–2 weeks
I sit with your team, map where the hours actually go, and rank every candidate workflow by ROI. You keep the map whether or not we continue.
- Shadow real work, not org charts
- Storefront ↔ ERP ↔ 3PL flows mapped end-to-end
- Ranked automation shortlist
- Build-vs-buy vs leave-alone calls
Build & Deploy
2–6 weeks
Agents, copilots, and automations built against your systems — in the cloud, or on a local model server I stand up on your hardware when privacy demands it.
- Right model per job: frontier API or local open-weight
- Evals so quality is measured, not vibes
- Amazon · Shopify · WooCommerce ↔ 3PL integration chops
- Docs and runbooks included
Team Enablement
ongoing
Your people learn to drive the tools themselves: prompt craft, agent habits, quality checks. The goal is fluency — so you never need me for the basics.
- Role-specific playbooks
- Working sessions, not webinars
- Internal champion coaching
Retained Operator
monthly
A forward-deployed AI operator on call: monitoring what we shipped, tuning it as models improve, and taking the next workflow off your list each month.
- Fixed monthly scope
- Model upgrades handled
- Direct line — no ticket queue
03
Meet your agent.
The terminal up top is the engine room. Here's one way the work could look from your team's seat — an illustration, not a product.
No console. No prompt engineering. Your people describe the job the way they'd describe it to a new hire — and the agent works through the steps while they watch.
Pick a scenario and watch it run. It's a sketch, not a product — the interface gets shaped to fit each business — but the steps are the real shape of the work.
- Plain-English tasks, not commands
- Every step visible and auditable
- Result lands where your team already works
task
04
How an engagement runs.
01
Listen
Start with a free 15-minute intro call, or book a working session straight away. Either way, you show me the messiest part of your operation and I tell you honestly whether AI belongs in it — and what I'd leave alone.
02
Prototype
Days, not months. A working agent against a slice of your real data, in your team's hands within the first two weeks. We kill what doesn't earn its keep.
03
Deploy
Into production with evals, fallbacks, and a human-in-the-loop where it matters. Measured against the hours and error rates we baseline in week one.
04
Hand off
Runbooks, training, and your team driving it. Then we pick the next workflow off the audit map — or you take it from here. Both are wins.
05
Built for 10–50.
That size is the sweet spot, and it's underserved on purpose.
Big enough to bleed hours
Enough people doing repeatable knowledge work that automation pays for itself in weeks, not quarters.
Small enough to move
No procurement theater, no six-month sign-off chains. The person who says yes is usually in the room.
Ignored by the big firms
Enterprise consultancies won't take your call; SaaS vendors just sell you seats. You get a senior operator instead.
One person, end to end
You deal with the person doing the work. When something breaks at 9am, it gets fixed by 10 — by the person who built it.
06
The operator.
For the last three years I've lived in this stack — daily-driving Claude, GPT, Kimi, GLM, Codex and Claude Code, running a local LLM server on my own hardware, and building agent runtimes by hand. Not reading about them. Operating them, every day, on real work.
Before that: ten-plus years inside enterprise systems — Oracle Cloud ERP, order-to-cash, procure-to-pay, and the integration glue between Amazon, Shopify, WooCommerce and 3PLs. So when I walk your floor, I already know where the hours bleed — exception queues, mismatched orders, swivel-chair re-keying — because I've run those systems from the inside.
That matters because the difference between an impressive demo and a dependable system is a thousand small judgments: which model for which job, where an agent needs a human checkpoint, when a local open-weight model beats a frontier API on cost and privacy. Those are decisions you can only make if you run the tools yourself.
This website was designed and shipped end-to-end by an AI agent under my direction. That's not a gimmick — it's the working model I bring to your business.
07
Asked, answered.
Our data is sensitive. Does it have to go to an API?
Honest answer: for most teams of 10–50, a tightly scoped API is the right call — we define exactly which fields touch it, and redact the rest. If you truly need on-prem, I can stand up an open-weight model on your own hardware and teach your team to run it (a consumer desktop handles lightweight models fine) — and I'll tell you plainly when that's overkill.
Models change every month. Won't this rot?
Systems are built model-agnostic with eval harnesses, so swapping Claude for GPT for a local model is a config change, not a rebuild — and you can prove quality didn't regress before switching.
What does this cost?
$200/hr, after a free 15-minute intro call. That's the whole pricing page: working sessions, audits, and enablement are all billed at the same rate; builds are quoted as a fixed hour budget per workflow, agreed up front. AI tooling isn't cheap and neither is hard-won judgment — the rate covers both. No retainers, no surprise invoices, ever.
Will my team actually use it?
That's the whole point of deploying with your team instead of at it. Prototypes go into real hands in week two; anything your people won't touch gets killed early, cheaply.
Are you going to sell us a platform?
No seats, no licenses, no lock-in. I build on your infrastructure with mainstream tools, and the hand-off package means you own everything outright.
Our world is ERPs and integrations, not chatbots. Does AI touch that?
That's where it pays first. Ten-plus years of Oracle Cloud ERP, OTC/P2P, and storefront ↔ 3PL integration work means I start where your exceptions live — order mismatches, ASN delays, invoice disputes. Agents are very good at triage, matching, and drafting inside exactly those flows.
08
One working session.
Zero slideware.
Bring your messiest workflow. In one hour I'll tell you whether AI belongs in it, what it would take, and what I'd leave alone — even if the answer is "not yet."
Free 15-min intro call · $200/hr after that · Replies within one business day