CS + Business Double Degree @ Waterloo & Laurier

Building cloud services, market data systems, and AI developer tools

📍 Toronto, Ontario, Canada

I study Computer Science at the University of Waterloo and Business Administration at Wilfrid Laurier University. I work as a Software Engineer Intern at RBC Wealth Management, building cloud services and data-access tools for AI agents. My projects include streaming market data systems, AI code review tools, and LLM fine-tuning.

• 🏦 Current work: AWS and OpenShift services, REST and GraphQL APIs, and a migration to official Snowflake SDKs.
• 🔭 Recent project: Prediction Markets Order Flow, with a C++20 order-book engine and durable market data capture.
• 🔍 Seeking: SWE / ML Engineer Co-op opportunities for Winter 2027.

• Order-book engine: Reached 234K updates/s at 1,000 levels per side, 12× the Python dictionary reference throughput in a local benchmark.
• Live capture: Observed 127 book states/s across 8 Polymarket markets and 16 books in a live capture run.
• Data pipeline: Built durable SQLite capture, transactional Kafka workers, sequence-gap recovery, and a FastAPI/React dashboard.

• Plan to code: Built a desktop tool that turns ChatGPT Web research and planning into structured Markdown handoff files for Codex CLI.
• Workspace access: Exposed Git state and files through 5 FastMCP tools over Tailscale Funnel, scoped to one granted workspace.
• Execution controls: Added permission checks on every call, path-traversal blocking, and single-use local approvals tied to SHA-256 plan hashes.

• Model training: Fine-tuned DeepSeek-Coder 6.7B on 2,400 curated problems via QLoRA.
• Evaluation: Increased pass@1 from 24% to 34% on a fixed 100-problem execution benchmark; hard-problem pass@1 reached 28.6%.
• Local Deployment: Built an evaluation harness with sandboxed execution, merged LoRA adapters, and exported to GGUF for local Ollama inference.

• Parallel review: Cut review time 2.2× with 3 AI agents reviewing in parallel and a moderator that checks findings against the diff, filters false positives, and merges duplicates.
• Desktop interface: Built an Electron macOS app with React and TypeScript, showing findings by severity beside code diffs.
• Review controls: Added Python thread-pool fallbacks for agent or moderator failures and human approval before GitHub CLI posts comments.

• Market scanner: Built a pandas and NumPy scanner over 500+ S&P 500 and Nasdaq 100 tickers using yfinance data to detect fair value gaps and equal highs and lows.
• Historical simulation: Recorded a 74% simulated win rate across 700+ historical setups on NVDA, GOOGL, and AAPL.
• Dashboard: Deployed a React/TypeScript interface with TradingView charts and WebSocket updates through a Flask API.

• Team delivery: Led sprint planning for a 4-person team across 4 sprints, building a Kotlin Android app with Jetpack Compose, Gemini meal analysis, and Supabase authentication.
• Meal analysis: Reduced response variability by 40% on repeated meal photos through Gemini temperature tuning, JSON-mode responses, and client-side retries.
• Meal logging: Reduced logging time by 20 seconds per entry versus manual input, with structured calorie and macro fields, persistent sessions, and real-time sync.

Watch the demo

Toronto, ON · Sept 2026 to Present

• Service performance: Reduced p95 latency by 70% during peak load for AWS and OpenShift wealth services through autoscaling, load balancing, and Redis rate limiting shared across clusters.
• AI data access: Built CRM tools serving 30,000+ requests/day through REST and Apollo GraphQL APIs in Node.js and TypeScript, with Redis-cached authentication tokens.
• Platform maintenance: Leading the migration of custom Snowflake integrations to official Snowflake SDKs.

Toronto, ON · Jan 2025 to Apr 2025; Sept 2025 to Dec 2025

• Data pipelines: Cut Python pipeline runtime from 4 hours to under 15 minutes for 500K+ listings on AWS through parallel workers and Redis caching.
• Service reliability: Cut failed-request investigation time by 26% with correlated JSON logs, exponential-backoff retries, and circuit breakers.
• Reporting: Replaced manual weekly reporting for 20+ product stakeholders with Python ETL, PostgreSQL, and Tableau dashboards.
• Content scoring: Ranked opportunities across 20K+ articles using scikit-learn regression and cohort analysis.

Toronto, ON · Jan 2024 to Apr 2024

• Data consolidation: Saved 300+ engineering hours/year by combining 4 databases into one AWS data layer with automated sync across Dockerized Python and Node.js services.
• User engagement: Increased active users by 40% in a Seattle Kraken NHL rewards pilot with 200+ users through a SendGrid notification platform with templates for each user segment.