Systems we've shipped, in production.

Detects Asian hornets at a bait station and tells individually marked insects apart, so a single hornet can be followed between visits.

Marked and re-identified individual hornets tracked at the bait station by their paint-dot colour

• Python
• YOLO
• OpenCV

Segments structures in a brain MRI volume and renders the labelled regions across all three standard viewing planes.

Three planes sagittal, coronal and axial views segmented from the same volume

• Python
• PyTorch
• Segmentation

One phone-grade camera turns a racket-sport session into a timeline of every shot, bounce and wall-target hit.

1 camera broadcast-style shot, bounce and speed tracking. No rig, no wearables

• Python
• PyTorch
• YOLO

A QC station that already knows every brand’s points of measure and tolerances. The operator just measures.

Every piece measured to each brand’s tolerance table, logged to the operator

• Next.js
• TypeScript
• Python

Real-time docking assistance: detect the vehicle, track the approach, log the turnaround, all from one fixed camera on the bay.

1st place IEEE Hackathon win, then delivered to a client

• Python
• YOLO
• OpenCV

A forward-facing camera segments the track bed, rails and ballast in every frame for automated track-condition monitoring.

Frame by frame track bed and ballast segmented from a forward-facing camera

• Python
• YOLO
• Semantic segmentation

A retrieval-grounded assistant that answers from verified internal documents covering pricing, tax rules and policies, and shows its sources.

Grounded answers every response cites the internal document it came from

• Python
• LangChain
• FAISS

Real-time detection and counting of confectionery on a moving conveyor, with every piece assigned an ID and counted exactly once.

Counted once each piece tracked across frames, even when it overlaps

• Python
• YOLO
• OpenCV

Object detection and tracking count cakes on a bakery production line from a single overhead action camera.

Live tally cakes counted off an overhead GoPro as they move down the line

• Python
• YOLO
• OpenCV

AnomalyCLIP flags breaks, gaps and weave irregularities in tire-cord fabric from a line camera, drawn as a live heatmap.

Flagged live breaks, gaps and weave irregularities called out as the fabric runs

• Python
• AnomalyCLIP
• Deep learning

A Detectron2 instance-segmentation model outlines and classifies vehicle damage from a single photo, with a review UI.

Per-panel damage segmented and classified from a photo of the vehicle

• Python
• Detectron2
• Faster R-CNN

A lane-detection pipeline that works on raw LiDAR point clouds. No camera, and robust in the dark and in glare.

3D lanes lane lines extracted directly from the LiDAR point cloud

• Python
• LiDAR
• Point cloud processing

State-of-the-art anomaly detection (ANOMALIB) for manufactured parts, deployed to run on edge hardware at the line.

On the edge defect detection running on-device at the station, no cloud

• Python
• ANOMALIB
• PyTorch

An OCR and extraction pipeline that turns scanned documents into structured fields, with a review view for low-confidence pulls.

Structured unstructured documents in, typed fields out

• Python
• OCR
• OpenCV

Computer vision on existing site cameras gives a live occupancy map for parking and a headcount for foot traffic.

Live map bay-by-bay occupancy and headcount from existing cameras

• Python
• YOLO
• OpenCV

Mission planning and simulation for an autonomous quad, with waypoints and behaviour validated in Gazebo, then flown through QGroundControl.

Sim to flight missions planned and validated in Gazebo before they fly

• Gazebo
• QGroundControl
• PX4 / ArduPilot