Computer Vision AI Projects: Real-Time Video and Edge AI

SyncTech is a Sydney property intelligence platform for insurers and claims teams. I lead R&D on its 3D computer vision technology.

I was building the technology engine for a new agriculture venture designed to put capital and modern agronomy into the hands of 1000+ rural agripreneurs in Kenya and India: an AI-first architecture from scratch to scale to thousands of farms, data-driven digital agronomy pipelines powered by LLMs and machine learning, and predictive farming with multi-agent systems to select the right lands and crops.

I helped shape Lima Labs' computer vision products for highly accurate yield forecasting and analytics on commercial farms in Kenya. By shifting away from complex mobile dashboards to visual insights displayed on central TVs, we drastically improved adoption and helped field teams spot trends and yield shifts earlier.

I advised a heavy equipment manufacturer on an edge-computing CV system to monitor worker productivity and ensure safety across heavy machinery plants. The pipeline tracks tool usage, operation cycle timing, PPE compliance, and danger zone intrusion detection to drastically reduce downtime and safety violations.

I designed a comprehensive Vision AI pilot concept to automate packaging and internal logistics for dry pet food production. The pipeline visually tracks the product journey from "bag to truck", featuring real-time seal quality inspection, 360-degree pallet scanning, forklift tracking, live inventory mapping, and dispatch mismatch alerts to prevent loading dock errors.

I built, in three months, a sub-second computer vision MVP that turns a live stream of several game windows into a structured game log in about 0.35 s per frame, with YOLO11n, PaddleOCR on TensorRT and OpenCV on a cloud GPU.

As a Computer Vision Advisor for an AI startup focused on deep facial data, I provided ML/Deep Learning expertise and initiated product development.

I provided product design consulting and helped build the R&D team for a high-speed sports analytics platform capable of processing live tracking at 120 FPS for professional coaching and sports markets.

I designed a computer vision system that reads waste bin fill levels from cameras, so urban waste operators can plan collection routes around bins that are actually full.

I designed and built a real-time proof of concept on a working drilling rig: CCTV and rig sensors feed an edge server that spots risky operations, such as a drill string left without movement, and alerts the crew on the spot.

In a weekend agritech hackathon, our team built a working prototype that finds fields prone to waterlogging from Sentinel satellite imagery (NIR, SWIR1 and RED bands): U-Net segmentation of the tiles, in a web app where users select an area and export the result as GeoJSON. Eight teams entered; ours reached the final four.

I helped design a computer vision system for dairy farms that used advanced pose estimation to identify individual cows. The pipeline continuously monitors their health states and rumination periods, turning physical farm activity into structured health data.

A Kinect goalkeeper game for Danone and FC Zenit: players stand in front of a screen and save penalty kicks with their bodies. The game ran on stands in Moscow and Saint Petersburg.

My first computer vision project: an interactive game stand for Nestea (Coca-Cola) in Moscow shopping malls, where players catch falling ice cubes with their bodies. We used Xbox Kinect cameras, custom drivers and OpenCV pose estimation, made the stands, and later rebranded the game for Coca-Cola.