Skip to content
View RyanJHamby's full-sized avatar
13 followers · 4 following

Block or report RyanJHamby

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
RyanJHamby/README.md

I make AI compute fast, efficient, and reliable, from the serving layer down to the hardware and the power behind it.

Software engineer at AWS (New York). Rust, C++, and distributed systems.

Projects

  • GPU Flight Recorder: finds the straggling rank in distributed GPU training and explains why it is slow, by joining PyTorch NCCL collective timings with per-GPU NVML telemetry (Go, gRPC). Validated in simulation; real-GPU run next.
  • FlowState: Rust as-of join engine with zero-copy Arrow, Rayon-parallel merge scans, and streaming watermark joins.
  • Order Book Engine: C++20 matching engine; P50 0.21 µs, P99.9 3.1–3.2 µs on EC2 c6i.large, lock-free SPSC ingestion, slab memory pools.

Open Source Contributions

Also built

Pinned Loading

  1. distributed-gpu-training-flight-recorder distributed-gpu-training-flight-recorder Public

    Finds the straggling rank in distributed PyTorch training and explains why it's slow, by joining NCCL collective timings with per-GPU NVML telemetry. Go agents, gRPC coordinator.

    Go

  2. tokenomics-bench tokenomics-bench Public

    Cost and energy per token of LLM serving under a p99 SLO: vLLM batching, prefix cache, FP8, CUDA graphs, and GPU power caps, measured together.

    Python

  3. FlowState FlowState Public

    High-throughput, Rust-accelerated temporal join engine for quantitative finance ML pipelines. Engineered via PyO3 to eliminate look-ahead bias and accelerate point-in-time feature generation on mas…

    Python

  4. order-book-engine order-book-engine Public

    High-performance C++ order book: P50 0.21μs, P99.9 5.7μs, 2.4M orders/sec. SPSC lock-free queue, thread-local slab allocator, price-time priority matching. 51 tests, EC2 Spot benchmarking.

    C++ 1

  5. stock-screener stock-screener Public

    Automated stock screener scanning 3,800+ stocks daily for confirmed Stage 2 uptrends passing 8 strict criteria. Features: smart caching (74% fewer API calls), market regime filtering, automated sto…

    Python 55 45

  6. macro-factor-decomposition macro-factor-decomposition Public

    C++20 macro trading system: PCA regime classification on 8-factor surprise covariance, Kelly sizing, IB Gateway execution. Sharpe 0.28, Calmar 0.65, 3.2% max drawdown on 17-year FRED backtest.

    C++ 1