Work on AI, Machine Learning, Computer Vision, and Large Language Models (LLMs) related projects. Mostly about model development, model optimization, and model deployment using different optimization engines and strategies.

- Develop proof-of-concept (PoC) of Machine Learning Operations (MLOps) for computer vision tasks and data pipelines using Flyte and SuperGradients.- Conduct research and evaluation of State of The Art (SOTA) models foundation models for medical imaging and Visual Question Answering using Vision Language Models (VLMs).- Develop and maintain guidelines to learn and understand the process of model development from data processing, model development, and model deployment using various… Show more - Develop proof-of-concept (PoC) of Machine Learning Operations (MLOps) for computer vision tasks and data pipelines using Flyte and SuperGradients.- Conduct research and evaluation of State of The Art (SOTA) models foundation models for medical imaging and Visual Question Answering using Vision Language Models (VLMs).- Develop and maintain guidelines to learn and understand the process of model development from data processing, model development, and model deployment using various tools. The repository is accessible in Kalbe Digital Lab Github.- Developed training pipeline for vision models for breast cancer classification using Lightning and W&B. Deploy the pretrained model for public showcase using Gradio in Hugging Face Space.- Develop and maintain public repositories for educational purposes and public demonstration in Kalbe Digital Lab Github and Hugging Face. Show less

In short, I'm responsible to develop deep learning model computer vision models and maintain core system modules. Some of them are:- Responsible for modifying, converting, optimizing, and deploying models to production using ONNX and TensorRT.- Develop and maintain Python modules for QlueVision system to interact with deep learning models and other functionalities.- Research, develop, and collaborate with team members to build machine learning applications for Indonesian Vehicle… Show more In short, I'm responsible to develop deep learning model computer vision models and maintain core system modules. Some of them are:- Responsible for modifying, converting, optimizing, and deploying models to production using ONNX and TensorRT.- Develop and maintain Python modules for QlueVision system to interact with deep learning models and other functionalities.- Research, develop, and collaborate with team members to build machine learning applications for Indonesian Vehicle Detection, Face Liveness Classification, License Plate Detection, and Road Pothole Detection.- Adapted NVIDIA Vision Programming Interface (VPI) into QlueThermal camera pipeline to reduce the computation load in the main system by utilizing CUDA computations written in C++.- Developed an AutoML platform for a virus monitoring system using Dagster scheduling system. Show less

- Assigned to collect, prepare, and train object detection models to detect road potholes. The final model was able to achieve 73% of accuracy.- Collaborated with backend and frontend to develop an AI application to analyze road potholes from submitted videos. The application was designed to process multiple videos at once using a parallel processing method and the model was deployed using Triton Inference Server.

- Created a hospital-like environment using Gazebo Simulator for mobile robot simulation.- Collaborated with other researchers to develop mobile robot software using Robot Operating System (ROS). It has the capability to navigate in a complex environment and avoid obstacles while moving around. The behavior was developed using Flexbe Engine to create tasks and sequences of actions. The navigation system was provided by ROS libraries.

- Developed the software system and interfaces using Robot Operating System (ROS) using C++- Developed the vision system using image processing and deep learning methods to detect balls and field lines for localization.- Contributed to the development of the walk engine for bipedal robot based on Inverted Pendulum and Zero Moment Points (ZMP).