Experience

Patterned AI. London, UK

Back on the team after my StartUp Lab term.

• Own Motif Studio end to end, a feature that splits a repeating pattern into separately editable motifs, from first prototype to launch.
• Built the cloud processing on Modal: a FastAPI service that runs each upload as a background job on hosted GPUs, with progress tracking, cancellation and per-job isolation so one failed job never affects the rest.
• Built the image pipeline in three stages: a segmentation model outlines every motif, a refinement model cleans up the edges, and the Gemini API identifies each motif and groups related ones.
• Raised motif edge accuracy from 74.5% to about 92% by fine tuning the segmentation model and adding a post processing step, specifically desgned for edge refinement.
• Built the React and TypeScript editor for the separated motifs with custo toolings.

• Python
• FastAPI
• PyTorch
• SAM 3
• Modal
• TypeScript
• React
• Gemini API

StartUp Lab. Waterloo, ON

• Owned the student side of a three-sided marketplace end to end, from data model to UI, for 1,000+ users.
• Scaled weekly client processing 20×, from 100 to 2,000 clients, by engineering an AI-driven pipeline.
• Designed the internal CRM staff use to manage students, employers and projects, with role-gated access to sensitive data.
• Built a nine-stage application lifecycle into the student dashboard in React, with real-time status updates.
• Drafted the platform's HECVAT and PSIA security and privacy assessments, documenting how personal data is handled across the stack.

• TypeScript
• React
• Next.js
• Gemini API

Patterned AI. London, UK

• Shipped Tile To Pattern, a full-stack Canva app, in three weeks. It passed 1,500 installs in its first month and grew to 15,000+ monthly active users.
• Cut the company's release cycle 5× by designing task-specific Claude sub-agents, skills and workflows.
• Built an image processing pipeline, curating over 150,000 images to enhance dataset integrity to improve AI model training
• Scaled the image pipeline to 30,000+ images an hour across 10 parallel workers, over 10× its previous throughput, and cut failures 95%, by targeting identified issues.

• Python
• TypeScript
• React
• Canva Apps SDK
• Claude Code
• MCP

WatArrow, UW SAE Aero Design. Waterloo, ON. watarrow.com

• Built and run the team's internal platform on Directus, Postgres and Redis with SSO. It manages members, purchases and sponsorships, and powers watarrow.com.

• Directus
• DigitalOcean
• Postgres
• Redis
• SSO

Wat Street, UW Quantitative Finance. Waterloo, ON

• Engineered the graph-attention module in PyTorch for a GAT+LSTM model of how volatility spreads between S&P 500 stocks.
• Designed the Pandas pipeline that turns 1.9M+ rows of S&P 500 data into structured training inputs.

• PyTorch
• NumPy
• Pandas
• Graph Neural Networks