CS + Business Double Degree @ Waterloo & Laurier

Building at the intersection of AI, Finance, and Scalable Platforms

📍 Thornhill, Ontario, Canada

I am a Computer Science (BCS) and Business Administration (BBA) double degree student at the University of Waterloo and Wilfrid Laurier University. I build production-grade data pipelines, fine-tune large language models, and ship full-stack financial and developer tools end to end — from model training to deployment.

• 🔭 Currently building: AI agents, LLM fine-tuning pipelines, and quantitative trading tools.
• 🌱 Currently learning: distributed training, retrieval-augmented generation, and low-level performance optimization.
• 🔍 Currently seeking: SWE / ML Engineer Co-op opportunities for Winter 2027.
• ⚡ Fun Fact: I fine-tuned a 6.7B parameter LLM on my laptop and Kaggle just to solve LeetCode problems faster.

• Agentic Handoff Portal: macOS desktop + CLI bridge that lets you plan on ChatGPT Web and execute with a local CLI agent (Codex, Claude Code, Antigravity).
• MCP over the Wire: Exposes a local Model Context Protocol server securely over Tailscale Funnel so remote chat sessions can drive on-device tooling.
• Token Efficiency: Offloads heavy planning to the browser frontend, reserving local agent context for execution.

• Performance Leap: 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; targeted data curation increased hard-problem pass@1 from 9.1% to 28.6%.
• Local Deployment: Built an evaluation harness with sandboxed execution, merged LoRA adapters, and exported to GGUF for local Ollama inference.

• TUI Dashboard: Built a terminal interface orchestrating local AI agents for automated PR code reviews.
• Non-Blocking UI: Used asyncio workers so the interface stays responsive during long AI review cycles.
• Security First: Integrated native gh CLI auth (no raw API keys stored) and staged reviews locally so comments are never posted without explicit confirmation.

• Real-time Scanning: Vectorized pandas engine detecting institutional support levels (FVGs, equal highs/lows) across 500+ tickers.
• Backtesting Engine: Custom simulator evaluating 700+ historical setups across NVDA, GOOGL, and AAPL, hitting a simulated 74% win rate.
• Dynamic Visuals: Deployed with real-time TradingView chart rendering backed by a yfinance data pipeline.

• AI Nutrition: Gemini vision pipeline converting meal photos into structured nutrition estimates with schema-constrained output and retry handling.
• App Architecture: Android app built with Jetpack Compose and MVVM, with asynchronous state management through Kotlin Coroutines.
• Gamified Retention: Forest gamification system syncing user progress to a PostgreSQL backend.

Toronto, ON — Sept 2026 – Dec 2026

• Cloud Platforms: Engineering an interactive financial forecasting engine and an AI-driven advisor assistant using React, microservices, AWS, and OpenShift.

Toronto, ON (Hybrid) — Sept 2025 – Dec 2025 · Jan 2025 – Apr 2025

• Data Scale: Scaled AWS Python pipelines to crawl 500K+ URLs, cutting runtime from hours to minutes via parallel workers + Redis caching.
• Stakeholder Analytics: Built weekly ETL into PostgreSQL and Tableau dashboards used by 20+ leadership and marketing stakeholders.
• ML Modeling: Built a scikit-learn scoring engine ranking 20K+ articles to prioritize content strategy.

Toronto, ON — Jan 2024 – Apr 2024

• Database Consolidation: Dockerized AWS ETL pipelines merging 4 siloed databases, saving 300+ engineering hours/year.
• Notification Engine: Built a SendGrid pipeline with queued retries for an NHL Kraken rewards pilot, sustaining 80%+ weekly engagement.
• User Segmentation: Built a pipeline clustering 100+ user profiles by behavior for personalized rewards delivery.