Hi! I'm Sanjeew Kanagaraj,
a Software Engineer based in Hong Kong

• Acted as Product Lead for the Advanced Driver Assistance (ADAS) team; interfaced directly with end-users to define features and translated needs into technical roadmaps
• Led the ADAS team in designing and developing patented advanced driver assistance algorithms including virtual anchor, heading hold, and 360° joystick control; currently deployed on a wide variety of vessels in 15+ countries
• Redesigned and optimised firmware architecture written using ROS2, C++ and Python, boosting runtime efficiency by 40% and reducing memory usage by 30%
• Developed a prototype of an autonomous navigation system for boats with minimal hardware, using a combination of traditional methods and reinforcement learning techniques achieving path accuracy of over 80%
• Created custom CI/CD pipelines with integrated Gazebo simulation for automated software testing and deployment, achieving 80% test coverage and reducing manual testing time by over 50%

• Developed multimedia pipelines to enable live video streaming in 4K, recording, two way intercom and deep learning inference using Gstreamer, Python, and Typescript, reducing processing and memory requirements by 80%.
• Trained and optimised computer vision models to perform various tasks including mask, gate and intruder detection, achieving over 90% accuracy on limited hardware using PyTorch, ONNX and OpenVINO.
• Improved obstacle avoidance and self docking success rate by implementing a graceful controller for charger detection and docking at firmware level using C++.

• Developed computer vision models for real world use cases using Tensorflow and Pytorch, optimised and deployed inference on edge devices using Docker, TensorRT and Nvidia Deepstream SDK.
• Implemented real-time face mask detection in conjunction with temperature checking, with precision of over 95%.
• Set up machine learning pipelines using Kubernetes and Kubeflow to streamline data pre-processing and model creation, improving efficiency by 40%.

• Co-authored paper titled A Neuro-Symbolic Humanlike Arm Controller for Sophia the Robot, researching the use of Convolution Neural Networks coupled with symbolic AI for object grasping. (Link)
• Constructed kinematic URDF representation of robotic arms to be used in simulation of intricate movements.Implemented AI behaviour tree algorithms to enable complex behaviour patterns on Sophia, integrating with the Hanson Robotics SDK.
• Improved Sophia’s human-robot interaction heuristics and reduced interaction delay by 50%.

• Worked on the coral reef mapping drone that automates the mapping of coral reefs.
• Responsible for training and deploying an onboard neural network to classify coral types. Built using Tensorflow and Keras
• Assisted in modifying electronic components to ensure accurate mapping and stitching of images.

• Worked directly under the CTO as Full Stack developer for the Quokka Rewards platform.
• Responsible for developing a Django RESTful API as well as data analytics to show the platform's effectiveness.

The University of Hong Kong

• Worked under Dr. Joe C.H. Yuen to build a cloud-based AI chatbot that helps students learn how to code.
• Responsible for full-stack development and deployment of the chatbot, which analysed hundreds of student submissions for common mistakes.