Data & AI Analyst | UBC MBAN | Forecasting · Optimization · ML

I build AI agent workflows and data products: two MCP servers published on PyPI (a tabular-data profiler and a Canadian capital-gains engine), a multi-agent investing framework where Buffett, Munger, and Burry personas debate my portfolio, and the forecasting, optimization, and machine learning work below. Claude Code and MCP are built into how I work every day.

• Shipped data-profiler-mcp and acb-tax-mcp: on PyPI and the official MCP Registry, listed in awesome-mcp-servers, Glama score A for both
• Building claude-multi-agent-investing: teaching 12 agents to argue about my portfolio so I don't have to
• Contributing fixes upstream to the tools I depend on: darts, PuLP, sec-edgar-mcp, yahoo-finance-mcp (see Open Source below)

12 agents (Buffett, Munger, and Burry personas, 6 fact-gathering analysts, 1 portfolio-manager synthesizer) debate every holding through 13 slash commands: tax-aware for Canadian accounts, advisory only, zero API cost beyond a Claude Code subscription.

MCP server that lets an LLM understand any tabular data file (CSV, Parquet, Excel, JSON): schema, per-column statistics, severity-ranked data-quality flags, and memory-saving dtype suggestions. pip install data-profiler-mcp, CI-tested on Python 3.10 to 3.13.

Canadian adjusted cost base and capital gains for LLM agents: CRA average-cost method, per-disposition realized gains, superficial-loss detection, unrealized gains against market prices, and a normalizer that turns raw broker CSV exports into clean transactions. Decimal-precise to the cent, zero heavy dependencies, a calculation aid (not tax advice).

Forecasting 14 years of monthly energy consumption for Vancouver International Airport with Box-Cox transformation and seasonal ARIMA: 1.7% MAPE on holdout data, with 3-year projections supporting procurement planning.

Two-stage ML pipeline (logistic regression screening, then random forest diagnosis) that cuts unnecessary lab tests by 30% while keeping 95% detection sensitivity (F1 = 0.87).

Mixed-integer linear programming model reassigning 2,000 students after a school closure: base solution $1.73M/year with 29 buses, plus a weather-constrained scenario quantifying the safety vs. cost trade-off.

More projects: US stock price prediction (VAR / LASSO / LSTM) · 10-K NLP industry analysis · TikTok reviews sentiment analysis (Python + R) · Nutrition planning optimization (LP)

Besides maintaining data-profiler-mcp and acb-tax-mcp (both on PyPI), I contribute fixes upstream to the tools I depend on:

• unit8co/darts #3167 optional min_train_length on local forecasting models, so short time series can train models that previously hard-required 10 points (merged)
• coin-or/pulp #936 expose CBC's randomSeed on COIN_CMD for reproducible MILP solves (under review)
• sec-edgar-mcp #142 fix 9 tools whose descriptions were silently empty (f-string docstrings never reach __doc__), with a regression test (under review)
• yahoo-finance-mcp #16 start/end date-range support for historical prices (closes an open issue)
• yahoo-finance-mcp #17 US class-share ticker normalization, so BRK.B stops silently returning empty data

• Languages & Data: Python, R, SQL (PostgreSQL, BigQuery), Pandas, scikit-learn, PyTorch
• Modeling: time series forecasting, optimization (MILP / LP), statistical inference, ML pipelines
• BI & Visualization: Power BI, Tableau, Looker Studio, advanced Excel
• AI & Developer Tools: Claude Code (MCP, CLI, API), GitHub Copilot, prompt engineering, AI workflow automation

• 2026-07-30 Published release in haiiibin/claude-multi-agent-investing
• 2026-07-29 Published release in haiiibin/data-profiler-mcp
• 2026-07-29 Published release in haiiibin/acb-tax-mcp
• 2026-07-28 Published release in haiiibin/data-profiler-mcp
• 2026-07-28 Published release in haiiibin/acb-tax-mcp

• Master of Business Analytics (MBAN), UBC Sauder School of Business (Nov 2025)
• BA in Economics, Minor in Commerce, UBC (May 2024), graduated with distinction
• CFA Level II Candidate

Happy to talk AI-agent tooling or analytics work. Reach me here: