researchlearning ml, writing it downclick again to enter →passionsgym · mma · gaming · readingclick again to enter →projects17 systems, built rawclick again to enter →worku of a · grad 2028click again to enter →connectsay helloclick again to enter →
the paperlearning cv from the pixels upthe oracle bugsfour failures, none in the modelthe allocator bugthe block too small to freethe tombstone bugthe delete that did not deletethe order book buga use-after-free that never crashedthe raft timeout bugwhen the runtime pauses too long
lifting6x a week, push/pull/legsmmatwo-time national championgamingapex plat · valorant diamondbuilding thingsallocators, vms, raytracersapollothe myth behind the nameall 7 →reading, basketball, the rest
oracle_of_delphia fully-local voice assistant. five rust services, a 14b model on my own gpu, nothing leaves the box.nexus_clusterraft consensus from the paper: elections, log replication, a replicated key-value store.nexus_dbembedded LSM key-value store in C++17, with a live telemetry dashboard.nano_matchlow-latency limit order book and matching engine.cpu_rasterizera 3d rendering pipeline in plain javascript, no graphics api at all.all 17 →the full cluster
computing science · university of alberta
abir deol
apollo
I work on the Vulkan validation layers and GPU drivers, and build systems from the bottom up: consensus, storage engines, allocators, virtual machines and rasterizers.
- 14 merged upstream
- Khronos · Mesa · ggml-org
- 10 live demos
or click a star to explore the void. you can also draw the bowstring and let go.