Bobby Mathews Blog Writing
Rust/wgpu systems engineer. I build GPU-accelerated real-time visualization for high-frequency sensor data.
About
I build GPU-accelerated real-time visualization for high-frequency sensor data in Rust and wgpu. Physical sensors and instruments produce continuous data streams faster than CPU-rendered software can display: the universal bottleneck is architectural, not domain-specific. The fix is the same across domains -- samples to a ring buffer, single per-frame GPU upload, signal processing in compute shaders, all channels rendered in one draw call, inside a 16ms frame budget.
My background spans the full systems stack this rests on: production async Rust (Tokio backends, embedded no_std), GPU compute (CUDA, wgpu/WGSL), and high-throughput data pipelines where latency and throughput numbers are the deliverable. I work from first principles and validate empirically.
Featured Work
A reusable Rust/wgpu engine for real-time visualization of high-frequency, multi-channel sensor data. Ring-buffer ingestion, single-draw-call rendering of hundreds of channels, and domain-specific signal processing in compute shaders -- all inside a 60fps frame budget, native and in-browser via WASM/WebGPU.
Target performance:
256-384 channels at 1-2kHz, 60fps · GPU compute pass running concurrently in the same frame
Architecture:
Lock-free ring buffer -> single per-frame vertex buffer upload · WGSL compute shaders for
signal processing · single draw call per frame · cross-platform via wgpu (Vulkan/Metal/WebGPU)
Rust · wgpu · WGSL · WASM/WebGPU
Other Projects
Production real-time data pipeline: lock-free hot path, exactly-once semantics, sub-200ms ingest-to-query latency, WAL crash recovery, and gap detection with RPC fallback. The high-throughput streaming and latency-discipline work that the visualization engine's ingestion layer builds on.
Rust · tokio · tonic · axum · Solana / TimescaleDB · PostgreSQL
Controlled benchmark isolating Tokio executor starvation under blocking workloads. Key finding: 101x p99 latency increase under starvation. EC2-controlled environment, reproducible results.
Rust · criterion · tokio
Research system for constrained document verification. Formal grammar verification runs client-side in WASM; local inference experiments use WebGPU. Technical substrate: finite-state parsing, browser deployment, and constrained AI workflows.
Rust · WASM · WebGPU · pest · logos · petgraph
Spot instance orchestration prototype for GPU inference. EC2 tag-based worker discovery, interruption-aware draining, stateless failover. Worked end-to-end; not pursued for economic reasons.
Rust · AWS EC2 · vLLM · SSM
Writing
I write about deep technical topics in real-time GPU rendering, high-frequency data systems, and infrastructure.
Expertise
Core competencies for real-time visualization and high-performance systems:
Contract Work
I take focused engagements building real-time visualization and GPU compute systems for teams working with high-frequency sensor or instrument data -- research instrumentation, robotics, medical and scientific devices, and industrial monitoring.
Typical scope: a focused rendering or signal-processing module -- ring-buffer GPU renderer, compute-shader signal processing, cross-platform (native + browser) deployment -- with a defined deliverable you can evaluate.