Projects¶

Applied AI Systems¶

LocalMind-Core-GX10¶

Enterprise multi-agent AI platform on DGX-class hardware. Sole developer. A polyglot architecture — a Rust performance core for dispatch and caching, a NestJS + TypeScript API gateway, and Python ML services — with a unified provider router and an agent layer over LightRAG hybrid retrieval (vector + keyword + knowledge graph). One platform replaces ad-hoc per-team integrations, with built-in tracing, regression testing, and safety validation. Python · TypeScript · Rust · vLLM · LightRAG · NestJS · DGX (SYNCROBOTIC, Sep 2025 – Present)

Research / Privacy-Preserving ML¶

Federated Learning Lab¶

From-scratch federated learning — FedAvg / FedProx / SCAFFOLD, DP-SGD and secure aggregation, plus FedPer / Byzantine-robust / FedAdam / FedLoRA. 33/33 tests, literature-cross-validated, with honest negative results on Non-IID MNIST. Python · PyTorch · Federated Learning · Differential Privacy waynehacking8/federated-learning-lab Notes: Federated Learning & Differential Privacy

NVIDIA Inference & Multi-GPU Stack¶

A portfolio of reproducible harnesses on the NVIDIA-native stack, run on 4× H100 (NVLink) and an RTX Pro 6000 (Blackwell). Each repo states what it is, what it is NOT, credits the upstream tools it drives, and ships measured results rather than claims.

GPU Inference Benchmarks & Kernel Studies¶

LLM serving internals, hands-on: KV-cache behaviour, quantization trade-offs, Flash Attention; hand-written kernels (tiled GEMM, WMMA Tensor Core matmul) profiled with Nsight Compute to connect kernel-level decisions to end-to-end serving latency. CUDA · Triton · PyTorch · Nsight (in progress — code release planned)

TensorRT-LLM + Triton Multi-GPU Serving¶

Reproducible build → serve → benchmark harness: TensorRT-LLM engines on Triton (TP=4), measured head-to-head against vLLM under matched concurrency. TensorRT-LLM · Triton · vLLM · 4× H100 NVLink (release planned once results land) Notes: serving on TensorRT-LLM + Triton

NCCL Collectives Benchmark¶

Bus-bandwidth micro-benchmarks for all-reduce / all-gather / reduce-scatter on 4× H100 NVLink, analysed against the theoretical link budget. Measured: all-reduce 366 GB/s (77% of NVLink); NVLS (NVLink SHARP) > Ring > Tree; protocol study (Simple/LL128/LL). NCCL · nccl-tests · CUDA · 4× H100 NVSwitch waynehacking8/nccl-collectives-bench Notes: where TP inference hits the NVLink wall

NIM Agent Blueprint¶

Agentic RAG reference architecture on NVIDIA NIM microservices (LLM + embedding + reranker) with a plan → retrieve → generate → validate loop and a built-in eval harness. Measured: retrieval recall@3 94%, and 0% hallucination on out-of-corpus questions with a guarded prompt vs 50% without it (ablation). NVIDIA NIM · RAG · agents · OpenTelemetry · FastAPI waynehacking8/nim-agent-blueprint Notes: 0% vs 50% hallucination

Blackwell Tensor Core Kernels¶

Hand-written CUDA GEMM kernels (naive → tiled → WMMA Tensor Core), benchmarked across Hopper (sm_90) and Blackwell (sm_120) as a fraction of the cuBLAS ceiling. CUDA · Tensor Cores · WMMA · Nsight · H100 / RTX Pro 6000 (release planned once results land) Notes: CUDA Tensor Core GEMM (WMMA)