Backends where losing state costs money. And the systems underneath them, rebuilt from scratch.
Six years on backends where the failure modes are expensive: data-intensive pipelines at the United Nations, a high-concurrency on-chain game economy at Crypto.com, now leading the platform behind a 200K-MAU self-custody wallet, the layer that makes many unreliable chains behave like one dependable thing.
One thread runs through all of it: reliability and correctness when something costly is on the line. Money, uptime, state you can't lose.
By night I go a layer down. I read the source of the systems I lean on (Redis, the Linux networking stack, database internals) and rebuild the interesting parts from scratch in C and Rust. "It works" and "I understand it" are different sentences.
- Solana, to the metal: how validators hold consensus, how the runtime executes a transaction
- Low-level Rust, the kind where the lifetimes fight back and you learn why
- Reading real systems and rebuilding them: event loops, storage engines, network servers, one primitive at a time
side projects, written to learn, not to ship.
Systems & storage, in C
- bytekv, a small Redis: kqueue event loop, append-only persistence, fork-based compaction
- C_Networking, a TCP server five ways, blocking → kqueue, ending in a framed protocol with backpressure
- RowOrient_DB, a tiny in-memory database, queries as function pointers
Rust & Solana
| Keyboard | Corne split, 42 keys, Gallium layout |
| Editor | Neovim · LazyVim · hardtime.nvim on strict (no arrow keys, no mashing hjkl) |
| Workspace | Zellij panes inside Ghostty |
Small, sharp tools, wired together by hand. First principles all the way down to the keys.