Skip to content
View imReese's full-sized avatar
🎯
Focusing
🎯
Focusing
5 followers · 6 following
  • Beijing
  • 04:55 (UTC +08:00)

Achievements

Achievement: QuickdrawAchievement: Pull SharkAchievement: YOLO

Achievements

Achievement: QuickdrawAchievement: Pull SharkAchievement: YOLO

Block or report imReese

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. 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
imReese/README.md

Hi, I'm Reese.

I work where inference engines meet real clusters.

AI infrastructure · inference systems · distributed runtime · performance engineering

Website Blog Email

About

I'm a software engineer on the Training and Inference Engine team at Baidu AI Computing, working on large-scale LLM inference deployments.

My work centers on inference runtime performance, attention-state storage and reuse, and parallel deployment. I optimize request scheduling and asynchronous execution, including overlap between cache transfers and GPU computation.

I contribute to our team's Attention Store, supporting the Kimi, DeepSeek, and GLM model families, including KDA state handling and L3 storage integration for pipeline-parallel inference. My runtime work includes prefill acceleration, disaggregated serving, and constrained decoding.

Selected Work

Building modular, engine-neutral components across inference systems, attention architectures, and runtime platforms:

Target Architecture

Design direction across these projects. Dashed arrows show planned interfaces; implementation details live in each repository.

Target architecture: Locus handles request orchestration and compute placement; NexusKV handles state identity and reuse planning; runtimes connect with external storage and transfer backends.

Tier Project Focus & Highlights Stack
Control Plane Locus Engine-neutral inference control plane for request orchestration and compute placement, informed by model-state locality. Rust Axum
State & Cache NexusKV Engine-neutral model-state reuse platform for state identity, compatibility-aware matching, and reuse planning. Go Rust Python
Engine Runtime sglang-rs Rust-native inference runtime exploring CPU scheduling, cache control, and asynchronous GPU execution. Rust gRPC
Attention & Kernels parallax-lab Hands-on experiments for evaluating attention architectures, associative recall benchmarks, and peak memory usage. PyTorch CUDA Python
Quant Platform Karkinos Local-first quantitative research and investing platform for China markets, with reproducible backtesting, risk controls, and paper validation. Python FastAPI React TypeScript
Engineering Notes imreese.github.io Personal site and source-level systems engineering notes with interactive components and deep dives. Next.js React MDX

Recent Notes

Toolbox

AI Serving & Runtime

SGLang vLLM Runtime Scheduling P/D Disaggregation KV Cache Hierarchy Prefix Cache & State Plane

Storage, Transfer & Infra

Linux Kernel & perf gRPC Mooncake Engine RDMA / RoCE Zero-Copy I/O Inference Control Plane

Languages

C++ Python Go Rust Bash

Earlier systems work
Before Baidu, I worked on cloud workload characterization and CPU architecture at Huawei Cloud's Shuhai Lab, and on distributed-storage control-plane systems at Huawei Data Storage.

Activity

GitHub contribution snake

Beijing, China · building systems that stay understandable under load

Pinned Loading

  1. Locus Locus Public

    Locus is an engine-neutral inference control plane for compute and model-state placement.

    Rust 1

  2. NexusKV NexusKV Public

    NexusKV is a model-state intelligence layer for inference systems.

    Python 2

  3. Karkinos Karkinos Public

    Local-first quantitative investing platform for reproducible research, evidence-bound decisions, risk controls, paper/shadow validation, and human-supervised execution.

    Python 1

  4. sglang-rs sglang-rs Public

    Rust runtime work for SGLang, covering scheduling, prefix caching, CUDA execution, and P/D disaggregation.

    Rust 1

  5. parallax-lab parallax-lab Public

    Hands-on experiments for implementing and evaluating attention architectures.

    Python 1