Systems and AI tooling projects focused on reproducible benchmarking, memory behavior, robotics integration, and practical developer products.
| Project | What it demonstrates | Stack |
|---|---|---|
| KV-cache Consumer GPU Benchmark | Measures KV-cache footprint, latency, throughput, and OOM boundaries across cache strategies on consumer GPUs. The documented RTX 5080 sweep contains 80 cases: 76 successful runs and 4 OOM cases. | Python, PyTorch, CUDA, Hugging Face |
| lab-tools | Reproducibility-oriented benchmark framework with CPU, CUDA, and Apple Silicon profiles; provenance manifests; energy collection; acceptance bundles; and statistical reports. | Python, Shell, CUDA, C |
| Calen | macOS menu bar calendar planner integrating Google Calendar, Apple Calendar, Reminders, and locally installed Claude Code or Codex CLI. Distributed through a Homebrew tap and GitHub Releases. | Swift, SwiftUI, EventKit |
| Azas | Seven-person ROS 2 cocktail-robot project using a Doosan M0609, RG2 gripper, and RealSense camera, with staged no-motion and hardware safety gates. | ROS 2, Python, TF2, MoveItPy |
| Memory Hierarchy Experiment Framework | C microbenchmarks and perf automation for cache locality, access patterns, TLB-related behavior, and matrix blocking experiments. |
C, Linux perf, Python, Shell |
- AI inference systems: KV-cache memory pressure, long-context performance, and GPU measurement
- Systems performance: memory hierarchy, reproducible experiments, energy, and provenance
- Product tooling: native macOS workflows and local AI integrations
- Robotics integration: perception-to-motion boundaries, staged verification, and operational safety
Python · C · Swift · CUDA · Shell · ROS 2 · Linux perf · GitHub Actions
- Financial Database — daily-updated public SEC EDGAR and OpenDART financial-statement dataset
- GDB Mentoring — 13-unit C/C++ debugging curriculum with examples and exercises
- Solved.ac profile