projects/

A native OCaml DSL for writing high-performance GPU kernels (GEMM, convolutions, etc.) targeting NVIDIA CUDA via NVRTC. Tesserae lets you express kernels in a typed, functional style, then compiles them to PTX and launches them on the device without leaving the OCaml runtime.

A high-performance large language model inference engine built in Rust with custom CUDA kernels. Axiom implements the LLaMA and DeepSeek model families with support for grouped-query attention, Mixture-of-Experts with speculative pre-gating, and LoRA adapter hot-swapping. Also consists of a ForkKV disaggregated KV cache that separates shared base prefixes from per-agent residual caches

cellc is a container runtime from scratch in C

Seraph is a kernel library for ARM neon assembly, along with a benchmark pipeline

Lattice is a Query Engine in C++