Curriculum vitae

Major: Engineering Mathematics, Statistics and Finance · Minor: Artificial Intelligence Engineering

Selected coursework: Stochastic Calculus, Mathematical Programming, Algorithm Design, Financial Optimization Models

Four-person research team supervised by Professor Ing-Haw Cheng

• Implemented and trained a single-market Deep Q-Network agent; contributed reward design, analysis, training controls, and PettingZoo environment extensions to team PPO and multi-agent market experiments.
• Independently designed reward and cost objectives and the controlled studies around them, then built the stopping, testing, analysis, and reporting workflow used to interpret agent behaviour across runs.
• Co-authored a four-author technical report, contributing sections on reward design and market-simulator evaluation.

• Founded and operate a multi-tenant communications platform connecting voice agents, workflow automation, calendars, customer records, and protected dashboards.
• HVAC Ontario’s owner attributes over $1.5M in annual revenue from Piston-handled leads and roughly $70K in yearly call-staffing savings to the system (customer-reported).
• Designed multi-agent workflows spanning calls, chats, email, analysis, and workflow optimization, with shared context, specialized roles, explicit tool boundaries, and human review paths.

• Worked directly with the fund manager across frontend, backend, and core-engine systems; added multi-agent routing for research workflows alongside financial-data validation, AI-assisted reporting, shared context, and authenticated usage and cost telemetry.
• The fund manager reported that the infrastructure reduced repetitive domain-specific analysis, broadened the instruments and asset classes the team could assess, and enabled analysts to make higher-level investment decisions. (reported)

• Led a DQN-versus-mean-variance portfolio-allocation project and developed an MIT-licensed Python backtesting package with 22 documented metrics, models, indicators, and analysis workflows.

Implemented robust ellipsoidal Sharpe optimization, sparse LASSO multifactor estimation, and a Sharpe-CVaR hybrid in CVXPY; the three-person team ranked 5th and 2nd of 10 groups across two held-out trials.

Owned classifier training and hyperparameter tuning for a four-person VGG-16 transfer-learning project; achieved 90.1% held-out accuracy, 7.9 percentage points above the optimized SVM baseline.

For a three-person analysis of 103,008 bookings, drafted methods, compiled results, and wrote the conclusion and limitations across transformations, diagnostics, influence analysis, and model selection.

Built an agent-facing Neo4j network for finding communities, connectors, and project-relevant people through hybrid semantic/graph retrieval with explicit evidence paths; exposed eight MCP tools with role-based filtering.