I build research-driven products across physical AI, resilient communication, computer vision, and quantum security. My work starts with difficult systems and ends with something people can understand and use. Explore the project notes below for the problem, technical approach, results, and links behind each build.
Muscle Memory keeps the MM-01 robot, its sensors, and its low-level walking controller fixed while apartment layouts and high-level task policies change. Every episode becomes replayable evidence that can shape the next curriculum without quietly rewriting the robot underneath the experiment.
Physical AI · 2026 · Co-builder and agent systems engineerSafeRelay combines a field relay and an operations console. The mobile app composes, validates, stores, and forwards emergency packets; the console reviews received evidence beside public hazard feeds. An optional Apple Foundation Models guide gives concise survival guidance on-device without turning model output into dispatch evidence.
Resilient systems · 2026 · Co-builder across mobile, console, and relay architectureZooVision turns fixed-camera footage into timestamped observations, keeps model evidence separate from policy decisions, and gives keepers one place to review the source clip, acknowledge an alert, record an outcome, and prepare a morning handoff.
Computer vision · 2026 · Co-builder and AI workflow engineerThis peer-reviewed simulation joins photon generation, fiber transmission, detection, quantum circuits, and configurable noise inside one end-to-end BB84 key exchange. It extends an earlier IEEE architecture into work published by Physica Scripta.
Physica Scripta research · 2026 · Co-first author and team leadAnchorMesh turns nearby phones into store-and-forward relay nodes. A compact distress signal can move over Bluetooth from one device to the next, survive intermittent connectivity, and appear on a responder dashboard only after a relay reaches the internet.
Disaster response · 2026 · Co-builder focused on protocol research, validation, and the responder dashboardMasterBuild starts with a raw product idea and dispatches specialist research agents across live online sources. Their findings return to one real-time workspace, where the system assembles a grounded go-to-market plan and an MVP direction that can be handed to Lovable.
Multi-agent systems · 2026 · Co-builder and multi-agent orchestration engineerTrainable connects robot registration and validation, evidence-backed physical assets, persistent 3D worlds, natural-language tasks, compatible simulation, versioned datasets, policy-training candidates, and measurable evaluation in one local-first workflow. It is built for robotics researchers, embodied-AI developers, and small robotics teams that need the path from an object or robot to a repeatable training run to stay inspectable.
Embodied AI · 2026 · Co-builder with Jerry across the embodied-AI workflow and systems integration