Computer science student at Miami University, graduating in May 2027. I build Python and TypeScript tools for agent memory, workflow controls, and evaluation.
| Project | Implementation | Evidence and scope |
|---|---|---|
| Multi-Agent Memory MCP | TypeScript, SQLite FTS5, MCP, typed graph retrieval | Tests, client examples, and a synthetic retrieval benchmark |
| AVP Tutor | Python tutor backend, execution snapshots, teaching-note retrieval, TypeScript client | Fork with upstream attribution, API tests, and browser tests with mocked model responses |
| CIFAR-10 CNN | PyTorch, ten weighted layers, configurable training launcher | Eight seeds averaged 96.84% selected validation accuracy. The eight-seed results contain no recorded test accuracy. |
| PMQS | Python replay, paper fills, settlement accounting, evidence checks | Synthetic examples and unit tests for research primitives |
Demo Tutor is a smaller Python API chat demo with one editable code example and draft teaching instructions. The page sends code and supplied output to the model. It does not execute student code.
Agent Workflow Control Plane contains task queues, model routing, memory interfaces, and typed authorization records. It is a development snapshot with documented test failures.
Event Contracts contains Python and Rust research, replay, and paper execution tools. Agentic Quant Operator records research decisions, paper evidence, and promotion refusals. Casino Line Modeling compares sports predictions with outcomes and market baselines. These research projects do not establish a profitable trading strategy.
The multi-agent system report records observations from one system. The project notes cover work whose source stays private.
I research decoder-guided agent recovery using entropy measurements and recent agent activity. I have prototype measurement and intervention code, plus exploratory analyses. The decoder needs implementation and a controlled evaluation.
The notes include an experiment plan, a proposed hidden-state probe, and work completed so far. I have no validated result showing that the proposed decoder improves agent recovery.
The setup guide links the public memory server, client examples, and three project-review skills. I use AI coding tools throughout my work, and I check changes against the project tests and records.