Production-grade automation and AI systems for content-heavy and delivery-heavy businesses. Built in weeks, not quarters.
I'm a builder who actually ran the ops — a decade of program management plus hands-on engineering. I take 2–3 clients at a time, ship working software every Friday, and leave your team running the system after I'm gone. No agency middle layer, no offshore handoff.
Every workflow runs through one human. They're great. They're also drowning. The work doesn't scale, and you can't hire your way out of it fast enough.
Editorial routing, client handoffs, QA cycles, capacity planning — the volume grew, the systems didn't.
You don't need another ChatGPT wrapper. You need production systems that route real work and don't break in week three.
Not strategy decks. Not workshops. Working systems shipped to production.
Routing, scoring, enrichment, and handoffs across your existing tools. The kind of work an ops coordinator does — running 24/7 instead of 9-to-5.
Claude and GPT pipelines that score, classify, summarize, and route — with eval frameworks so you know they actually work, not just look like they work.
Quality scoring, capacity planning, and SLA monitoring frameworks for teams shipping client work or content at volume.
For companies that know they need to move on AI but don't know where to start. Audit, prioritize, ship the first three production wins.
Anonymized stories — workflow problem, what changed, and the numbers after. Happy to walk through specifics on a call.
A two-sided marketplace was processing hundreds of publisher submissions weekly through manual review and routing. The ops team was the bottleneck for every transaction, and quality was inconsistent across reviewers.
A Zoom → n8n → Claude pipeline that captured intake calls, scored submissions against a structured rubric, routed approved publishers into the right marketplace tier, and flagged edge cases for human review. Eval framework wrapped around it to monitor scoring drift over time.
A multi-team services company knew they were behind on AI but had no internal capability, no roadmap, and no clear first wins. Leadership wanted real production deployments, not a pilot graveyard.
90-day audit and roadmap covering 12 candidate workflows. Prioritized three for immediate build: client deliverable QA, capacity planning, and contract intake. Shipped all three to production with adoption tracking and team training.
A team handling thousands of customer calls per week had no scalable way to monitor quality. Manual QA covered ~5% of calls; the other 95% were a black box.
An automated call scoring system using Claude to evaluate every transcript against a multi-dimensional rubric, surface coaching moments, and flag at-risk accounts in real time. Dashboard for team leads, weekly trend reports for leadership.
Three phases, fixed scope, working software at the end. Most pilots run 4–8 weeks end-to-end, sized to the workflow you bring.
We map the workflow that's hurting, the systems it touches, and what success actually means in numbers.
I send a written teardown of how I'd build it — architecture, integrations, edge cases, what's automated vs. left manual — and we agree on scope before any code is written.
2–6 week sprints depending on scope. Weekly Friday demos, working software at the end of every week, production deploy by the final sprint.
Three artifacts every pilot ships. Anonymized previews — the real ones come with your workflow, your systems, your edge cases.
Every step of the current process drawn out — owners, systems, handoffs, and where it breaks. The bottleneck stops being a feeling and becomes a diagram.
An 8–12 page written teardown of how I'd build it: triggers, actions, integrations, edge cases, what's automated vs. left manual. Reads like a real engineering doc, not a deck.
The actual LLM pipeline — prompts, eval rubric, confidence thresholds, and the human-in-the-loop boundary. Plus the eval harness so you can tell when it drifts and fix it without me.
My background spans program management at scale (PMP, CSPO, ITIL certified) and hands-on technical builds — production AI pipelines, automation systems, custom SaaS. I work best with founders and operators who want to ship real systems, not buy more software. I'm based in Toronto, take on 2–3 clients at a time, and care more about your team being able to run the system after I leave than about looking impressive on a deck.
Week 2 ends with a fixed scope, fixed price, and fixed milestone dates — signed before a line of build code is written. If I'm going to slip, you hear about it on the Friday demo, not the deadline.
Scope is locked at the System Design milestone. New asks go into a shared backlog and either swap for something equal-sized or become a small change-order. No silent additions, no surprise invoices.
You get a runbook, a 60-min training session with whoever owns it internally, and 30 days of bug-fix support included. After that, an optional light retainer if you want me on call — but the goal is your team running the system without me.
I build on tools your team can read — n8n, TypeScript, Postgres, Supabase — not bespoke frameworks. Every workflow ships with a runbook and an eval harness so you can tell when something drifts and fix it without me.
No. Discovery week exists for that. You bring the workflow that's hurting; I bring the teardown, the architecture, and a written recommendation. If it's not worth building, I'll tell you on the call.
There's no rate card. Pricing is scoped to the workflow you're trying to fix — a lean pilot for an early-stage team looks different from a multi-system rollout for a 200-person org. Tell me the constraint (budget ceiling, deadline, the smallest slice that would matter) and I'll come back with a pilot shape that fits — or tell you straight if it doesn't. Most engagements can be carved into a smaller first slice, phased rollout, or flexible payment cadence. The goal is an engagement you actually say yes to, not a quote you ghost.
30 minutes. No deck. You walk me through the workflow that's eating your team's hours, I tell you straight whether I can help.