Selected work
Experience
- May 2026 to nowShopifyApplied ML Engineering Internnow
I work on the LLM agents behind the shopping assistant people use in the app.
- Oct 2025 to Jul 2026PerfectiLead Engineer, Voice AI
Stayed on after the acquisition and took the voice platform from something that only worked in a demo to something 8,000+ mortgage brokers use.
- Jan to Mar 2026AnthropicClaude Campus Ambassador
Selected for the campus ambassador program, then founded the official Claude Builder Club at U of T. It’s at 450+ members now, and I ran five workshops plus a live build-along for about 200 students at CUCAI.
- Apr 2023 to Sep 2025Call FusionCo-founder
Started it in first year and ran it for two years before we sold. 150,000+ calls for 20+ clients by the end.
- Summer 2025HexawareSoftware Engineer Intern, AI/ML
Built two ML systems on my own, one that triages incoming support tickets and one that tracks head pose for remote exam proctoring.
About
What I build with
For coding I’ve used most of the harnesses out there. Cursor for a long time, then Claude Code and Antigravity, and lately Pi and Herdr, where I spawn subagents and agent teams and mix models across them based on how much reasoning the task needs. On the backend side it’s RAG pipelines, prompt and context engineering, evals and judge rubrics, and guardrails to keep a model in bounds. For fun I run an autonomous agent on a VPS that I talk to from my phone, with its own skills, cron jobs and routines. I was also an Anthropic campus ambassador, running workshops on Claude Code, MCP servers, agentic workflows and prompt engineering.
Now
It’s fall 2026. I’m at Shopify through December, and I’m applying for winter and summer 2027 internships in the US. I moved somewhere a fourteen-minute walk from the office this term, which has done more for how often I actually make it to the gym than any amount of planning did.
I just got back from Peru, my first trip off the continent with friends. Outside of work I read a lot about startups that failed and not much about the ones that worked, mostly because the failures are the ones you can actually take something from.