My work focuses on systems that make complex software workflows more reliable, inspectable, and usable by other people.
Deterministic runtime governance for autonomous coding agents.
An open-core platform that intercepts file operations from coding agents (like Claude Code and Aider), parses them with Tree-sitter AST, and blocks unsafe operations (secrets leaks, boundary violations, unauthenticated mutations) before code reaches disk.
- Tech: TypeScript · Python · Rust · Tree-sitter · JSON Schema · VS Code IPC
- Status: Open-Core Kernel (v0.1.1) · Monorepo Development
- Links: Repository · Website · Documentation · Quickstart Demo
Decision memory for coding agents.
A local memory harness that uses Git hooks to capture architectural decisions from commit diffs via LLM call, stores them in SQLite with vector embeddings, and re-injects relevant constraints into future agent sessions to prevent architectural drift.
- Tech: Python 3.11 · SQLite (C-runtime custom functions & recursive CTEs) · Git Hooks · BYOK LLMs
- Status: Distributed on PyPI (
v0.2.2) - Links: Repository · PyPI: klyd · Architecture & Workflow
A student-built coding platform for a student-run coding club.
A purpose-built competition and assignment platform running in school computer labs. Features client-side Python execution via Pyodide WASM, manual admin review workflows, and serverless SQLite atomic scoring with zero RCE attack surface.
- Tech: React · Vite · Monaco Editor · Pyodide (WASM) · Cloudflare Workers · Cloudflare D1 · WebCrypto
- Status: Internal tool deployed at PSBB Schools (~50–68 registered students in Grades 8–9 cohort)
- Links: Repository · Platform Information
A desktop cybersecurity demonstration combining ClamAV detection with explainable incident narration.
A standalone Electron desktop application built in a rapid 6-day sprint for Innovision 2026. Pairs real clamscan binary subprocess execution with high-speed Groq LLM inference to explain detected threats in plain English during live exhibit demos.
- Tech: Electron · React · TailwindCSS · Node.js Child Process · ClamAV · Groq API (gpt-oss-20b)
- Status: Solo-built exhibition demonstrator · Reported first-place result at Innovision 2026
- Links: Repository · Demo Context & Recognition
Existing systems lose important context, permit unsafe actions, or fail to support the people using them.
I tend to respond by building infrastructure:
| Infrastructure | Core Responsibility | Solves |
|---|---|---|
| WatchLLM | Governs agent write paths deterministically | Unsafe / unauthenticated code written by autonomous agents |
| Klyd | Preserves architectural decision lineage in Git | Architectural decay and context amnesia across agent turns |
| Codify | Supports practical programming education | Fragile manual lab coordination and unsafe student RCE |
| Sanboxx | Makes cybersecurity detection understandable | Inscrutable raw CLI antivirus output for non-specialists |
┌──────────────────────────────┬──────────────────────────────┐
│ Compilers, ASTs & Systems │ Cloud, Edge & Databases │
│ • Python (3.10+) · Rust │ • Cloudflare Workers & KV │
│ • Tree-sitter AST Parsing │ • SQLite / Cloudflare D1 │
│ • JSON Schema Draft-07 │ • Recursive CTEs & C-Funcs │
├──────────────────────────────┼──────────────────────────────┤
│ Agent Infrastructure │ Frontend & Desktop │
│ • Git Hook Interception │ • TypeScript · React · Vite │
│ • Deterministic Governance │ • Electron (Context Isolated)│
│ • Decision Memory & Vectors │ • Monaco Editor · Pyodide │
└──────────────────────────────┴──────────────────────────────┘
I also organize and teach through Codify, a student-run coding organization at school. We run weekly hands-on programming labs, prepare problem sets, and supervise live competitions to make computer science practical and accessible for younger students.