Pinned Loading

1. cuda-reranker cuda-reranker Public

   A hand-optimized CUDA attention kernel for RAG cross-encoder reranking, achieving ~25% lower attention latency than PyTorch SDPA on an NVIDIA T4.

   Python 1

2. overveckill overveckill Public

   A vector database in C++. Building the stuff I'd normally just pip install.

   C++ 1

3. cracked.c cracked.c Public

   From scratch neural network & autograd engine in C, arena allocator, PCG PRNG, unity build, zero dependencies.

   C 1

4. repo-trace repo-trace Public

   Corrective RAG assistant for exploring and debugging unfamiliar codebases, with retrieval grading, query refinement, reranking, web fallback, and citation guardrails.

   Python 1

5. train-maxx train-maxx Public

   Python

6. flash-attention flash-attention Public

   Cuda