Toronto, Canada · AI & Automation Engineer
Anyone can demo an agent. I ship ones that survive.
I own the architecture, the security review, and the bill — not just the demo.
What I do
- AI agent systems
- Multi-agent architectures that route work to specialists and hold up in daily production use.
- Automation & integration
- I can't watch a team do by hand what a script can do in a second — so I automate it.
- Backend built for failure
- I design for the restart, the timeout, and the user who configures something unreasonable — first.
- Cloud infrastructure & cost
- I make the bill visible and defensible, not a monthly surprise.
- Security in review
- I'm the one reading the code, and security is a comment on line 40, not a phase at the end.
- Frontend & prototypes
- Fast, rough, and real — built to get an idea in front of people before committing real engineering time.
The engine
"I go toward what I don't know yet."
How I get things done — the loop
01 / 06 — interaction: why-machine
Ask why
I don't build what you asked for until I know what it's for.
"Most of what makes a feature wrong is a wrong assumption nobody checked."
02 / 06 — interaction: kill-wall
Prototype fast
A hundred rough ideas, cheap. Rough and real beats polished and theoretical.
"Building the proper version first is how you spend three weeks on the wrong thing."
03 / 06 — interaction: kill-wall
Show it, kill it
"That's not what we want" is the most valuable sentence I can hear — in week one, not week four.
"Not building something is a real outcome."
04 / 06 — interaction: break-it
Build it and own it
I stay responsible after the deploy.
"Handing something off at the deploy step is where most AI features quietly die."
When the bill becomes an engineering problem
Audited and restructured cloud spend across multiple projects, made cost attributable and continuously visible instead of a monthly surprise, scheduled non-production resources down, and migrated and hardened access control. Ran a full infrastructure and source-control audit using an AI agent constrained to read-only access by construction — not by instruction.
The migration that unblocks itself
Led planning to move a product frontend off a third-party component library and onto an internal design system. The blocker wasn't the components — it was the developer experience of consuming an internal library as a dependency, and solving that is what actually unblocked the migration.
Most AI features die between the demo and production
A multi-agent assistant with a supervisor that routes work to six specialists, model selection across multiple providers chosen for cost and capability, a dual-layer memory system with conversation compaction, and streaming responses running over real network infrastructure — not just a local demo.
05 / 06 — interaction: spot-it
Secure it in review
Security is a comment on line 40, not a phase at the end.
"Telling a tool to be careful is not a safety design."
The security review that blocks the deal
Completed a full vendor security assessment of the kind that gates enterprise purchasing decisions, worked against a recognized security framework, and planned breach readiness — including what could actually be investigated with the logs already being kept.
06 / 06 — interaction: hand-counter
Automate the manual
I can't watch the team do by hand what a script does in a second.
"The manual work I spot is what sends me back to asking why it's done that way at all."
A manual process quietly eating weeks per cycle
Replaced a fully manual renewal process covering more than a thousand customer accounts. Connected billing, CRM, and spreadsheet data, automated quote generation, and built visibility into churn that didn't exist before.
An agent that does the homework before the meeting
An autonomous research agent that researches every company on an upcoming meeting list and delivers a structured, current brief beforehand, using search-grounded generation so the information doesn't go stale. Delivered straight into the chat and document tools the team already used, so adoption required no behavior change.
What I didn't build
Evaluated it against the actual requirement and it was more complexity than the problem needed.
Saved: Months of maintaining an abstraction nobody needed.
Researched it end to end and found the data quality didn't hold up for what it would have been used for.
Saved: Weeks of integration work on a source that wouldn't have delivered.
Shown to the team early — it wasn't the shape of the actual problem.
Saved: Three weeks of building the wrong thing correctly.
Started as a rough prototype, survived every round of feedback.
The arc
- Scope expanded into AI engineering
From QA automation into designing multi-agent LLM systems.
- Joined as a co-op student
Scoped to QA automation.
- Hired full-time
The scope started widening almost immediately.
- Took ownership of cloud infrastructure and cost
- Started mentoring the next co-op students
- Took on security and compliance work
Vendor security assessments, breach readiness, log strategy.
Things I had never done before I did them
- Building multi-agent AI systems
- Owning production architecture
- Owning a cloud bill
- Training the co-op he once was
- Leading a security review
18 months.