Pinned Loading

1. nanoinfer nanoinfer Public

   An LLM inference engine written from scratch: NumPy forward pass, Rust INT8 kernels, BPE, KV cache, sampling and quantization, plus a live browser view of its internals. No transformers or llama.cpp.

   Python

2. tinydelta tinydelta Public

   A single-node table format with an atomic JSON log, time travel, and optimistic concurrency.

   Python

3. toy-cpu toy-cpu Public

   8-bit multi-cycle CPU in structural VHDL: hardwired control, 4x8 register file, 14-instruction ISA. Synthesized for Cyclone V in 73 ALMs.

   JavaScript

4. nanocompile nanocompile Public

   A tensor compiler for the transformer nanoinfer runs by hand: graph IR with symbolic shapes, fusion, schedules and C codegen, checked against nanoinfer

   Python

5. RAPTOR RAPTOR Public

   From-scratch Autotune in Python/NumPy: STFT pitch tracking, equal-temperament note quantization, and modulation-based pitch correction. UVA Signals and Systems final project.

   Jupyter Notebook

6. leetcode-submissions leetcode-submissions Public

   My accepted LeetCode solutions, synced automatically from my submission history

   Java