Bobby M.

I build and fix high-performance backend and real-time systems in Rust, where latency, throughput, and correctness are the numbers that decide whether the product ships. When a service stalls under load, drops data, or cannot keep up in real time, I find the wall and remove it. I publish original benchmark research on async-runtime failure modes, so that reasoning behind a fix is legible, not a black box. What I do - Performance engineering in production Rust: diagnosing and removing tail latency, executor starvation, lock contention, false sharing, allocator pressure, and missing backpressure. These rarely show from the symptom alone. I find the real cause with tokio-console and flamegraphs and prove the fix with before-and-after p99 numbers. - High-throughput data and compute infrastructure: real-time ingestion and streaming pipelines with bounded end-to-end latency, exactly-once processing, crash recovery, and horizontal scaling with automatic failover. Hot-path in-memory work split from a cold analytical store so latency-sensitive processing never waits on historical queries. - Real-time systems with hard guarantees: lock-free SPSC ring buffers, single-copy handoff, no dropped samples, determinism treated as a requirement rather than a hope. - Rust and WebAssembly: compute and rendering compiled to wasm32 and running in-browser via WebGPU, native-class performance with no server round-trip. - GPU-accelerated processing and visualization of high-frequency sensor data: hundreds of channels at kHz rates with signal-processing compute (FFT, filtering, fusion) inside the same frame, native and in-browser. The hardest end of the real-time spectrum, where most CPU tooling gives out. - C, C++, Java and Python to Rust migration: the FFI boundaries, unsafe, and zero-copy interop that do not auto-port. Recent work - A high-throughput ingestion pipeline sustaining 40-plus events per second with zero drops: write-ahead-log crash recovery, gap detection with source fallback, and an auto-healing design built to scale past 5,000 events per second, with a hot in-memory path separated from a cold TimescaleDB store for historical and backtest queries. - Original research on async executor starvation showing a 101x p99 scheduling-overhead increase when blocking code saturates the runtime's workers, instrumented with tokio-console. When to hire me You have a domain-first product in instrumentation, hardware, infrastructure or cluster observability, robotics, sensing, or finance, and it is blocked by a hard systems problem outside your team's discipline, where getting the abstractions, the latency, and the correctness right matters more than shipping fast. That is the work I am built for. #Skills Rust, WebAssembly, Asynchronous Programming, Concurrent Programming, Systems Programming, Performance Optimization, Backend Development, gRPC, Microservices, Distributed Computing, Real-Time Data Processing, PostgreSQL, Apache Kafka, Linux, Embedded Systems, C++, C, GPU / Parallel Computing, Graphics Programming, API Development