Research
One question organizes my work: how do you represent a world — physical, simulated, or informational — so it can be understood and acted upon, by humans and by machines? I have pursued it through three threads: spatial computing (registration and tracking on head-mounted displays), distributed simulation (HLA-federated synthetic environments and their observability), and knowledge infrastructure (local-first systems that turn raw data into navigable structure). Machine learning is where these threads converge — representation made learnable — and is my current direction of study, with a longer-term interest in quantum information.
Projects, by research relevance
Local-first LLM workspace with autonomous entity extraction, persistent memory, and interactive knowledge-graph synthesis over conversation histories (Tauri/Rust + SQLite).
Seam-aware, content-adaptive 3D Tiles pipeline: METIS graph partitioning with seam-weighted cuts, crack-free attribute-aware simplification, and measured-error HLOD generation.
Flight-system analysis tooling for distributed simulation: observability and interoperability over live simulation state.
Reverse 6DoF tracking: synchronizing a moving physical platform's pose into a head-mounted display's spatial frame.
HoloLens spatial alignment: three-point calibration procedure registering holographic content to physical space.
Offline disk-space intelligence: parallel Rust filesystem scanner, persistent SQLite index, progressive XXH3 deduplication, and canvas treemap exploration.
Papers & technical notes
Working notes toward a paper on detecting aerial maneuvers from telemetry: a labeling suite over CesiumJS, a relational event schema, and a versioned feature pipeline for training honest baselines.
Working notes toward a paper on registering a virtual coordinate frame to a physical room from three measured correspondences — the similarity transform, its residuals, and where the naive solution breaks.
In preparation: a technical report on seam-aware content-adaptive tiling for 3D Tiles (void-tiler).