AI/ML engineer. I build LLM agents, big or small, virtual or physical, and I'm rigorous about evaluating them. My best work happens on teams that have fun doing it.
Washington St, Brooklyn
PC Jason Francis
Agents, evalsand edge models
Three co-ops shipping LLM systems, a research lab, and a hackathon platform that 200+ people used.
Waterloo, Ontario
Computer science at Wilfrid Laurier University, graduating December 2026. I work on agents, evaluation and small models that run where the data lives, and I research how robots and drones plan their routes.
Keep scrolling down. The path moves sideways.
Jun 2026 to now
Quantum ML Research Assistant
Swapping a DQN's classical value network for a parameterized quantum circuit in a multi-agent framework for decentralized drone navigation, and benchmarking returns and parameter counts against the classical baseline.
Jan to Apr 2026
AI/ML Software Engineer, co-op
Fine-tuned and benchmarked Qwen 2.5 3B on edge hardware so customer data stays on the device, and built a self-hosted OpenAI-compatible service that lets local models drop into existing LLM workflows.
May to Aug 2025
Machine Learning Engineer, co-op
Built a LangSmith evaluation system that runs LangGraph agents in parallel, and deployed the team's first custom MCP Excel server on Cloud Run with Terraform.
Sep to Dec 2024
Junior ML/AI Engineer, co-op
Wrote an auto-tester that scores retrieval quality, which lifted RAG accuracy 13% and ROUGE 6%, then reached 80% accuracy classifying user stories across domains and stacks.
Jan 2026 to now
President, Wilfrid Laurier University
The university's first AI/ML club backed by AI faculty: project teams, workshops that bring first-years into the field, research groups and hackathon teams.
Sep 2025 to May 2026
VP of Technology
Shipped the organizer control plane and the participant portal for a 200+ hacker event: constraint-based team matching, realtime QR check-in and a sponsor dashboard.
Three projects, each running live. Click and drag inside the boxes.
Jun 2026 to now
4 drones
A browser stand-in for the PyBullet environment: up to four quadrotors, each steering around moving pillars and each other on its own. The research trains a quantum-circuit policy for this; the demo steers with simple potential fields.
May to Sep 2026
nodes 0searching
Anytime RRT* growing live. The path gives the person room and keeps getting shorter. Click to move the goal.
IEEE, 2026
Scripted walkthrough with made-up scores: three agents analyze in parallel, the coordination agent fuses their confidence and decides. The real system reached 84.1% zero-shot detection.
May to Sep 2026
Ran a research RRT* planner as a drop-in Nav2 global planner, then taught it to bend around people and hazards. Found that stopping at the first solution silently switched off rewiring, and converted it to anytime RRT*.
23rd L&T Conference, IEEE, 2026
Three CrewAI agents, each on its own Mistral-small instance, analyze every email in parallel (BERT classification, WHOIS checks, threat intelligence). A coordination agent fuses their confidence and decides whether to quarantine.
Outside co-ops and research: platforms, agents and prototypes.
Next.js · Supabase · React · Flask
The software a 200+ hacker event ran on: an organizer control plane and a participant portal, open-sourced after the event.
On GitHub
Yellowknife, January 2026A dream, crossed off.
What I do when I'm not training models.
I graduate in December 2026 and I'm looking for full-time roles and internships.