Explore my portfolio ↗ · Email · Try NavoX ↗
I'm Michael, an Electrical & Computer Engineering student at Cornell University. I build high-performance C++ systems, ML infrastructure, developer tools, and reproducible research. What interests me most is making software both fast and explainable: clear invariants, useful tests, careful measurements, and a product people can actually run.
| Project | What I built | Start here |
|---|---|---|
| Ledger · C++23 systems | Allocation-free limit order book with price-time priority and lock-free SPSC handoff. 18.74M synthetic commands/s, 271 ns p99 on an AMD EPYC virtual host; 100M differential commands verified. | Design · Benchmarks |
| Gatehaven · C++23/SDL3 | Compiled circuit simulation, native desktop editor, binary saves, and downloadable macOS, Windows, and Linux development previews. | Try a preview · Build |
| NavoX · AI product | Personal assistant connecting inbox, calendar, and student workflows, with source-linked answers and review before external actions. Student-specific features are actively being developed. | Live site · Architecture |
| Fluxion · ML systems | From-scratch NumPy tensor engine, reverse-mode autodiff, GPT-style transformer, gradient checks, and optional C++/BLAS and experimental CUDA operators. | Architecture · Measurements |
| ToolRet · Retrieval research | Conditional reranking over 37,292 tools. On 1,500 untouched queries, the primary router made 23.4% fewer cross-encoder calls while meeting the preregistered nDCG loss tolerance. | Research paper |
| Disclosure Drift · Quant research | Source-audited SEC 10-K risk-disclosure study: 215 filings and 180 year-over-year comparisons. Descriptive pilot results are reproducible; the proposed return-forecasting study is not yet validated. | Methods & findings |
- Correctness first: independent reference-model checks, autograd tests, and explicit edge cases.
- Measurements with context: controlled performance workloads and held-out evaluation, including unfavorable results.
- Build it so people can use it: cross-platform Gatehaven previews, NavoX walkthrough, and runnable local examples.
My path into software includes Harvard CS50, Springboard software training, and hands-on programming practice. Since starting Electrical & Computer Engineering at Cornell in 2024, I've studied data structures, digital logic, computer systems, architecture, networks, databases, and machine learning. Those foundations inform the newer public systems and research repositories above.
I also maintain ModelForge (model versioning and canary deployment), MarketLab (portfolio analytics), Pro Competitive Programming (835 C++ reference implementations with local verification), and LeetCode Atlas (pattern-based Python/C++ interview practice).
Working with
