🏃♂️ Running enthusiast (newly unlocked) • 💻 Software & Data @ UofT • ☕ Powered by curiosity and an unhealthy amount of caffiene 📍 Toronto, Canada | 🎓 Computer & Data Science (Co-op, May 2027)
Recently got into running — turns out it’s great practice for facing problems head-on.
(Also great for debugging… eventually. 🥲)
I’m a Computer & Data Science student at the University of Toronto. I'm always trying to learn something new to help me approach problems with a new lens.
I enjoy:
- Taking ambiguous problems and turning them into simple systems
- Owning projects end-to-end (design → code → deploy → iterate)
- Learning by building (and occasionally breaking things responsibly)
Currently:
- 💼 Full Stack Developer @ Scotiabank (Global Banking & Markets)
- 🛠 Building ReturnFlow, a Rails returns & exchanges platform
- 🏃♂️ Running (mostly from bugs, sometimes from my responsibilities)
A self-serve returns portal + merchant dashboard designed like a real commerce platform.
Why I built it:
Because returns are messy, async, stateful, and full of edge cases — aka a great systems problem.
Highlights
- 🏬 Multi-tenant Rails 7 app (store-scoped data model)
- 🔁 Validated state transitions (requested → approved → received → refunded/exchanged)
- 🧾 Append-only event log for traceability & auditability
- ⚙️ Sidekiq + Redis for async jobs (notifications + webhooks)
- 🔐 Webhooks with HMAC signatures + retry/backoff
- 📊 Analytics on return reasons & SKU hotspots
Think Shopify returns, but built in a hackathon-style sprint (with fewer meetings).
📦 Repo: coming soon (currently sprinting 🏃♂️)
- Cut manual email triage by 80% by building an LLM-backed classification pipeline (RAG + evaluation + safe fallbacks)
- Reduced multi-team processing timelines from months → days by owning a mission-critical approvals platform end-to-end
- Built a cosine similarity REST service with caching + load testing to validate 10× scaling
- Reduced AWS containerization costs by 40% via multi-stage builds, right-sized compute, and worker consolidation
- Built a geospatial scenario builder UI (React/TS) for ML-driven planning with OpenAI + Ollama
- Reduced Django + PostgreSQL API latency by 35% via profiling, ORM optimization, and indexing
- Saved $12,000 by in-housing an event signup app serving 9,300+ users
- Cut Firebase reads by 30% via caching + query restructuring
- Reduced check-in time by 50% using SendGrid + QR-code automation
-
📈 Real-Time Market Simulation & Prediction Engine
Built a real-time FX market simulator with PyTorch; sped ingestion 6× using Polars over pandas. -
🌪 Hurricane Preparedness Aid
GenAI RAG over emergency documents + real-time tracking maps; Dockerized for reproducible deployment.
Languages
Python • TypeScript/JavaScript • Java • C# • SQL • Ruby (learning)
Frameworks & Tools
Rails • React • Django • Flask • PyTorch • Redis
Infrastructure
AWS • Docker • CI/CD • PostgreSQL
Data / ML
Transformers • RAG systems • Polars • Pandas • NumPy
- Recently started running — turns out consistency beats motivation (same applies to debugging)
- I like systems that are boring in production (the highest compliment)
- Favorite performance metric: “this used to take forever”
