Systems engineering for cameras, robots and edge AI.
MOAT Safety is James Isaac Kay's engineering practice in Texas. Bring a problem that involves sensors, embedded computers or autonomy. The work starts by writing down who it serves, what limits it, and how you will know it worked.
MOAT Safety is also the home of DashStar, a dashcam alert-matching research prototype.
What MOAT Safety does
Five kinds of work, each one something James has already built, flown or shipped.
-
Edge perception and video pipelines
Real-time detection, tracking and recognition on NVIDIA Jetson hardware, from the camera stream to a single clean event.
Done before: a 9-service DeepStream and Savant pipeline (YOLOv8n, LPDNet, NvTracker and LPRNet on TensorRT FP16) running 28 to 33 fps on recorded 1080p video on one Jetson Orin Nano. Track-keyed deduplication turned 2,286 raw plate reads into 1 event.
-
Sensor timing, calibration and fusion
Getting cameras, LiDAR and computers to agree on what happened and when, so fused data lines up while the platform moves.
Done before: GStreamer pipelines that stamp each frame's capture time into the H.264 stream (SEI NAL units and RTP header extensions). Field tests measured about 110 to 120 ms of video transport latency and a 35 s clock offset between hosts, closed with chrony and NTP, enabling camera-LiDAR fusion above 15 fps underway. ChArUco calibration at about 0.5 px reprojection error, and an anisotropic undistortion method for wide-angle dashcam lenses.
-
Autonomy, controls and flight testing
Control, guidance and test planning for uncrewed boats and drones, from simulation to supervised field tests.
Done before: FAA Part 107 pilot-in-command for autonomous flight tests under Safety-Review-Board-approved procedures. Co-developed vision-based UAV precision landing at 16.23 cm average error, and multi-mode PID and inverse-kinematic thrust allocation for an over-actuated WAM-V, validated in Gazebo VRX (SITL and HITL) and on the water. Migrated a vessel and its UAV from ROS 1 to ROS 2.
-
Requirements, reviews and release discipline
Stakeholder needs written down, design decisions recorded, and every release gated by tests with evidence. The habits that keep a small team from shipping surprises.
Done before: 13 tagged releases in 17 days, each gated by automated ship tests (300+ pytest and bash smoke tests) with recorded evidence. 41 design and plan documents, 35 recorded design decisions, and 31 plan errors caught by pre-build reconnaissance. Instrumentation with OpenTelemetry, Prometheus, Loki, Tempo and Grafana that root-caused a live-throughput ceiling. As a teaching assistant, mentors 27+ capstone engineers on INCOSE requirements, safety and design reviews.
-
Aerial survey and photogrammetry
Part 107 flight operations for industrial sites, with orthomosaics, 3D models and image-based defect analysis for equipment audits.
Done before: freelance aerial photogrammetry and 3D modeling of industrial plants since June 2025.
How an engagement starts
The first step is a video call at a time you pick.
-
Book a call
Pick a free 15 or 30 minute intro call and answer a few quick questions: name, email, phone, industry and what you would like to talk about. A calendar invite with a Microsoft Teams link follows.
-
Optional pre-meeting brief
After booking you get a short brief to fill in if you like: stakeholders, problem, constraints, success criteria, budget and timeline. Any question can wait for the call.
-
The call, then a needs brief
James uses the call to fill the gaps. Your answers become a one-page needs brief with open questions listed, and you get it to correct.
-
Proposal, or a plain no
If the work fits, a written proposal follows. If it does not, you hear that plainly.
The home of DashStar
DashStar is designed to read license plates in dashcam video on a small computer inside the vehicle and to check them only against official public alerts. As designed, when a plate matches, the device sends a cropped image of the vehicle and its plate (or just the plate), with other vehicles' plates and bystanders' faces blurred on the device, plus the location and time of the sighting. No video leaves the device. During the pilot, James is the reviewer at the MOAT Safety desk for every match. Vetted volunteer reviewers may help, but they see the cropped image only, never the location or call details. Only after a match is confirmed does the desk call the alert area's non-emergency number or the hotline named in the alert and pass on the location and time. DashStar Network will be free to join.
DashStar is funded by tips, sponsors and cloud credits, with crowdfunding planned for a lower-cost first camera. It will take no revenue until a pilot proves its accuracy.
DashStar is a research prototype for James's M.S. thesis. There is no live network yet, no real-world match has occurred, and accuracy has not been measured yet.
Six commitments DashStar is held to
- Plates that match nothing are discarded on the device within seconds.
- No location trail. Every match event carries the location and time of the sighting, because a sighting is useless without them. Nothing else ever carries a location, and drivers acknowledge this when they opt in.
- A human reviewer confirms every match before anyone is contacted. By default the driver hears only "do not follow". A driver who opts in can also see match notifications, marked unconfirmed, with the number to call. Drivers never follow or engage, ever.
- Footage is captured for training only if you opt in.
- Accuracy will be published, along with a quarterly transparency report.
- Matches use only official public alerts that include a vehicle or plate: AMBER, Silver, Endangered Missing and similar alerts, including stolen-vehicle alerts and suspect alerts such as Blue and Yellow alerts in the states that issue them. DashStar never takes private watch-lists or one-off plate requests.
Never:
- No data sales.
- No police-network integrations.
- No bounties.
Who you work with
James Isaac Kay is a perception and edge-AI engineer who builds real-time detection, sensor-fusion and video-timing systems on embedded NVIDIA hardware. They are an M.S. candidate in Electrical and Computer Engineering at Embry-Riddle Aeronautical University, expected to finish in November 2026, and hold a B.S. in Applied Mathematics from Southern Methodist University.
The MOAT work began in July 2025. MOAT Safety was registered in Texas as an assumed name (DBA) of James's sole proprietorship in July 2026.
- Team Minion, 1st of 20 international teams, 2022 Maritime RobotX Challenge
- Best Technical Design Report, 2024 Maritime RobotX Challenge (team)
- FAA Part 107 Remote Pilot
- Passed NCEES FE, Electrical and Computer (2025)