Rukon Uddin

Machine Learning Engineer

Aires

Feb 2025 - Present • 1 yr 5 mos

1. Led the development of a vision-based guidance system for visually impaired users at HK Airport, transforming it into a scalable landmark-based solution. Designed an archi- tecture supporting 20–25 concurrent users with only ~4GB VRAM using model pooling and queue-based processing. Improved navigation by integrating compass and IMU data, quantized models for mobile deployment, and built a low-latency WebSocket pipeline with FastAPI. Developed a simulation framework using synchronized video and sensor replay for remote testing and debugging. Additionally, implemented obstacle avoidance with Depth Anything V2 and rapidly prototyped an ARKit LiDAR-based obstacle avoidance system. 2. Built a LiDAR-based SLAM pipeline using FAST-LIVO2 for real-time 3D mapping and lo- calization. Worked with LiDAR, IMU, and camera sensor fusion, ROS/ROS2 bag processing, point cloud generation and visualization, trajectory estimation, and Docker-based deploy- ment for robotics and autonomous navigation systems. 3. Developed a locally deployed, FastAPI-based chatbot system designed for efficient on- device inference, capable of handling 20 concurrent users with scalability tied to avail- able compute resources. Integrated Celery for asynchronous task orchestration and background processing, ensuring responsive interactions and smooth workload distribu- tion while maintaining low latency in a resource-constrained environment. 4. Built the backend for a real-time translation system for smart glasses (SOLOS Glasses), deploying ASR and translation models on a local server and enabling low-latency commu- nication via FastAPI and WebSockets. Researched and handled the device-specific audio input format from the glasses’ microphone, implemented streaming of audio to the server for processing, and returned translated output back to the device, achieving a seamless end-to-end proof-of-concept for a Hong Kong museum use case.

AI Engineer

GenEQTY

Nov 2022 - Nov 2024 • 2 yrs

1. Developed and deployed a domain-specific chatbot in collaboration with the Robi team, utilizing RASA and LLaMA 3 for functionality and fallback handling. Implemented REST and SOAP APIs to provide features like balance inquiries, internet activation, and minute checks. Collaborated with other department members to create an effective chatbot interface. 2. Developed an Agent ChatBot monitoring system using Socket, enabling agents to mon- itor chatbot responses and intervene when necessary. Implemented RabbitMQ for efficient queue management. 3. Developed a highly accurate face recognition system with a 99.89% F1 score, reducing recognition time from 3.5 seconds to under 1 second. Implemented zkt API to open door when a recognized face arrives. The facial recognition system was trained on 100 employ- ees. Implemented a crowd monitoring system using IP cameras and built a dynamic crowd heatmap using DeepSORT and YOLOv5 for enhanced crowd management.