Structured-light scanner runtime
Rebuilt the acquisition pipeline for a structured-light 3D imaging system as a transport-agnostic Rust runtime — ~67 MB/s sustained, illegal state transitions caught at compile time.
Read case study →Mathematician & Systems Engineer
Systems Engineer
Trained as a mathematician, I build technical systems where difficult models, real-world constraints, and software must agree. My recent work is in 3D scanning and reconstruction, but I’m at home wherever abstraction, performance, and careful engineering matter.
I work primarily in Rust; though I've built production systems in C[++], CUDA, and Python. My focus is building research-grade algorithms, performance-sensitive runtimes, and production tooling with the occasional user-facing system.
3D vision systems, from sensor to surface
A selection of work across acquisition, reconstruction, optimization, embedded devices, and the infrastructure which connects them. Client work is described at a technical, client-agnostic level.
Production scanning systems
ICP registration, twice: CUDA and Rust
Outlier-robust point-to-plane ICP, built in CUDA and in generic Rust whose CPU and GPU paths compile from the same kernel source — frontend registration at roughly 3 ms per frame.
Read case study →Resilient graph-based reconstruction
SLAM built as graphs of direct observations end to end — no accumulated intermediate model — keeping reconstruction moving through tracking loss; trajectory error 3.16 mm → 0.76 mm, pose-graph optimization 10× faster.
Read case study →Owning the calibration engine
The solvers that turn images of a known target into the camera and pattern models everything downstream depends on — rebuilt around hand-derived analytical Jacobians, robust rejection, and bit-determinism.
Read case study →Accuracy metrology: decomposing a scanner's error budget
Built the instruments to measure how accurate a 3D scanner really is, then decomposed its error budget against industrial ground truth — down to a systematic scale error traced to the calibration target itself.
Read case study →Auditing the imaging front end, stage by stage
A numeric, stage-by-stage audit of three implementations of the same front end — closing parity to 98.9% of 1.1M centroids within a micron and isolating a one-sample filter bug worth 220 µm of depth.
Read case study →One core system for research and production
The Python binding layer that sketched the client's current C++ system, and the dependency-free C-ABI library that now delivers the Rust reconstruction core back into it.
Read case study →Independent work
mapr: a Rust library for 3D mapping and reconstruction
An independent R&D codebase for RGB-D SLAM, point-cloud registration, pose-graph optimization, and surfel-based reconstruction — the underlying geometry made explicit, testable, and reusable.
Read case study →entry: embedded systems for a findable, multi-credential key fob
A personal hardware and firmware project: an async secure-element driver, NFC-V bring-up cross-validated over RF and wired I²C, and compile-time ownership of every hardware connection surface.
Read case study →