Hanjun Yoo

• Developed a robotics AI agentic framework that finds the best sequence of subtasks for a given manipulation task.
• Designed the system to train a policy continuously without relying on heuristics.

• Developed a diffusion model that can generate interactive images, such as infinitely looping image or puzzle images by applying specific geometric conditions.
• Developed a robot manipulation policy that can better understand spatial locations, by integrating location information.

• Developed and improved document understanding model, using both data-based and model structure-based approaches.
• Added multi-language support (En, Kr, Jp, Cn), ability to distinguish Image/Molecule/Math Expression, and increased the maximum number of queries in prediction.
• Achieved 8%p performance increase for private testset, and scored 91% on publicly available testset.

• Developed a Knee-MRI analysis service that could assist medical staffs during diagnosis.
• Using 3 AI models, provides 1) The probability of this patient having illness 2) Importance of every slide in the MRI 3) Why the model thinks this way, using Grad-CAM.
• Achieved around 5%p higher sensitivity compared to the base research paper, and received positive feedback from doctors.

• A kaggle-like competition for multi-label segmentation of X-Ray images.
• Used augmentation methods that makes bone boundaries clearer, and combined instance segmentation results for better performance.

• Developed a service that detects and alerts users in forward-head posture.
• Alerts users who are too close to the screen, provides daily average distance to the screen, and provides hand-made stretch exercise videos.

Computer Science, Exchange student (In progress)

Computer Science, Exchange student

• TGPA 3.9 / 4.0 (4.38 / 4.5)

Bachelor of Computer Science (In progress)

• TGPA 3.9 / 4.0 (4.38 / 4.5)
• Member -> Division head of student council/union (Mar 2021 - Feb 2022)