I build things that extract structured data from systems that were never designed to hand it over: public APIs, scraped HTML tables, and raw game memory. Most of my work is in Python, C#, and Lua.
I'm currently focused on data analysis and working toward data science roles.
Battle of the Agents: Single vs. Multi · Python OpenAI statsmodels Plotly Quarto
The follow-up to the benchmark below. Having established that an agent beats a
plain LLM, this asks which kind of agent wins: one agent holding all three tools
against a three-role system — planner, solver, reviewer — over the same 100 real
search queries. Paired hypothesis tests and interactive plots across latency,
correctness, token cost and error recovery. The single agent turns out faster per
query and the multi-agent more accurate, with coordination overhead driving cost
up sharply. Read the report
LLM Agent vs. Baseline LLM · Python Groq statsmodels Quarto
Benchmarks an agentic LLM pipeline — planning, tool use, self-assessment, and
error recovery — against a single-shot baseline running on the same underlying
model, so agentic capability is the only variable. Every tool call is logged with
its latency and failure mode, then regressions measure how total latency scales
with task complexity, tool count, and API difficulty. Packaged as a reproducible
Quarto report that builds from committed results without calling a live API.
Historical Stock & Revenue Dashboard · Python pandas BeautifulSoup Plotly
Pulls full price histories for TSLA and GME through the yfinance API, scrapes
quarterly revenue from HTML tables with BeautifulSoup, cleans and normalizes both
into pandas DataFrames, and renders shared-axis Plotly dashboards comparing share
price against revenue over time.
Open-source tools for the Archipelago multi-game randomizer community — 27 tagged releases across three actively maintained projects.
| Project | What it does | Stack |
|---|---|---|
| Pokémon Snap Tracker | Reads live emulator memory to auto-track run progress in real time | Lua Python |
| Kirby Air Ride Tracker | Maps GameCube memory layout to a logic-aware progress tracker | Lua |
| Unfair Flips Tracker | Auto-tracking pack with item and location logic | Lua |
| Fable AP Mod Work | Reverse-engineered game data definitions to support another developer's project | C++ Python |
The common thread with the analysis work: find the signal in an undocumented format, validate it against known behavior, and package it so that people who didn't write it can rely on it.
Python · pandas · NumPy · statsmodels · Jupyter · Quarto · Plotly · BeautifulSoup · SQL · C# · Lua · Git
