Transformer internals: attention without the mysticism
A mechanical look at scaled dot-product attention, KV caching memory geometry, and rotary embeddings without the hand-waving metaphors.
Available for interesting problems
Building software, and writing about the road between.
I'm a software engineer working on distributed systems and developer tooling. This is where I keep my notes on technology, the places I've been, and the things I've built. No newsletter pop-ups, no cookie banner — just writing.
Recent writing
All posts →The Svaneti traverse: towers, mud, and Caucasian silence
Four days walking from Mestia to Ushguli through Georgia's high Caucasus. Medieval stone towers, glacial mud, and guesthouse hospitality.
Evaluating LLMs when benchmarks are gamed
Public leaderboards suffer from severe data contamination and metric hacking. How we design calibrated, private eval suites for production systems.
Selected work
All projects →raft-kv
A multi-package Raft consensus engine and key-value store in Go with snapshot compaction and a partition-injection test harness.
log-tailer
A fast structured log streaming CLI in Rust with zero-allocation predicate filtering, autodetection, and colorized formatting.
Get in touch
I'm happy to talk about backend architecture, developer experience, or where to eat in Lisbon. The inbox is open.
Say hello →