Backend Engineer
Backend engineer. I write Go (sometimes it compiles).
I think first, build second — so it tends to hold up in production.
Don't take my word for it. Look below.
play with the cube — drag, click, or pressdrag / tap /r u f
Git is opaque, so I rebuilt its core to see how it works. A working version control system from scratch in Go — init, add, commit, branch, checkout, status, log, and diff over content-addressed SHA-256 objects, with an LCS diff engine and a Bubble Tea TUI. ~1k lines, zero git libraries.
Background work that must not get lost. Distributed job queue for Go — PostgreSQL is the durable source of truth, Redis only the priority index, so jobs survive a full Redis wipe. Dead-letter management, embedded dashboard, CLI, and Prometheus metrics. Benchmarked at 52ns/op with 0 allocations.
Load testing that finds the real bottleneck. Fires HTTP, database, and Redis load on one clock, correlates latency across all three layers, and ships a shareable HTML report.
Stop polling the database. Minimal PostgreSQL logical replication client — streams committed row changes from the WAL with a five-line OnChange callback, webhooks, or a live SSE console. Decode kept pace up to ~37k changes/s with lag draining to zero.
Existing scripting languages felt either too heavy or too unreadable, so I wrote one. A scripting language in Go — Pratt parser, tree-walking interpreter, closures, goroutine-backed concurrency, sandboxed VM, and cross-platform binary compilation. Now the scripting layer for kyfram, a keyframe animation engine (linked below), and logos2d, a declarative 2D game framework.
JWT authentication library for Go. Rotating refresh tokens with theft detection via typed errors, and four pluggable storage backends: Redis, PostgreSQL, MySQL, SQLite.
Self-hostable feature-flag service backed by Postgres: flags live in your database and sync to flagd in real time — no polling, no extra platform, no vendor lock-in.
HTTP/1.1 server built from TCP sockets. Keep-alive, TLS, static file serving, graceful shutdown. Buffer pooling pushed throughput from 5,000 to 11,000 RPS.