NELSON
KANG
CS & Finance · Builder · UWaterloo
ABOUT
ME
I'm Nelson. I study CS and Finance at Waterloo and I mostly build backend systems: pipelines, async architectures, ML infrastructure. I'm happiest when the problem is poorly defined and the first challenge is just figuring out where to start.
When I'm away from a terminal I'm usually playing volleyball, running, or convincing myself that poker is a skill game and pretending I have an edge.
TECH
STACK
SKRYPT
HEALTH
AI Engineer at Skrypt Health, part-time and ongoing. Building lab-result triage on a live EHR, routing abnormal results into a staff review queue. Designed an autonomous coding harness, and fixed a production outage that had hidden every clinic support request since launch.
AQUANOW
Backend & Data Engineering Co-op at Aquanow. Ported a legacy reconciliation job to a Kafka-triggered Lakeflow worker over Apache Iceberg, then led its expansion into an AI-agentic workflow. Fixed 5 critical bugs in a live regulatory system reporting to Turkey's MKK — restoring reconciliation to 107/107 currencies.
ERICSSON
Software Engineer Intern at Ericsson Canada. Built 4 ETL pipelines in C# + SQL Server to extract Jira data, developed Power BI dashboards for trouble report visibility, and led a LEGO Mindstorms workshop on image-based sign recognition.
CLIP
FARM
Automated volleyball highlight generator. Users upload game film; YOLOv8-pose extracts player skeletons frame-by-frame, a rule-based classifier scores spikes, serves, and blocks, and FFmpeg re-encodes clips served via presigned Cloudflare R2 URLs. Audio RMS analysis reduces false-positive clips by ~20%.
VIEW ON GITHUB ›DOG
WATCH
2nd Place, Dominion Dynamics WHITEOUT Track at Hack the North 2026. Four machines — a quadcopter, a fixed-wing and two towers — search a 25 km Arctic strait for an unlit vessel and hold a track once found. A water-only probability field counts every empty camera sweep as evidence, so the fleet reasons instead of sweeping.
HUME
1st Place, CORTEX BioSciences Track @ Hack the 6ix. A deterministic belief-revision engine (no LLM) that ingests streaming evidence into a knowledge graph. Prompt-injection is blocked structurally — an independent ~90-attack red-team produced zero text-driven mutations.
STOCK
PORTFOLIO
ENGINE
1st place, Market Beat Competition. Multi-factor portfolio pipeline applying momentum, volatility, correlation, and trend signals (RAAM). Optimized a $1M CAD portfolio across U.S. and Canadian equities — 5.7% return vs. 3.7% benchmark, out of 18 teams.
VIEW ON GITHUB ›