Skip to content
View HANKSOONG's full-sized avatar

Block or report HANKSOONG

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
HANKSOONG/README.md

Hi, I’m Hank Song

I'm a recent MSc graduate focused on AI, with a focus on deep learning, computer vision, multimodal learning, and LLM-based agents.

I enjoy building AI systems that connect research ideas with working implementations — from computational imaging and representation learning to multimodal retrieval and tool-using agents.

I recently completed a year-long deep learning internship at Thermo Fisher Scientific in Eindhoven (2025–2026), where I worked on computational imaging, particle detection, and learning-based methods for challenging scientific imaging data.

Selected Projects

  • Charisma Predictor: Predicts personality & charisma scores via multi-modal fusion (video, audio, text). Achieved 92.5% accuracy using custom ensemble logic.
  • MiniVision: Benchmarks ResNet, EfficientNet, DINOv2 on CIFAR-10&100. ViT reached 98.7% & 91.5% accuracy.
  • Image Restoration: DnCNN vs. NAFNet on GOPRO/RealBlur with metric + perceptual analysis
  • qwen-quantization-benchmark: Benchmarking BF16 vs INT8 vs NF4 4-bit quantization
  • MiniGPT-TinyStories: A small GPT-style language model trained from scratch on the TinyStories dataset
  • Agentic-Multimodal-Product-Decision-System: An LLM agent that combines personalized multimodal retrieval, catalog search, and review evidence to make grounded product recommendations (continuously updated)

Interests

  • Multimodal Learning & Representation Learning
  • Computer Vision & Vision-Language Models
  • Large Language Models & Agentic AI
  • Uncertainty-Aware Machine Learning
  • AI Systems, Retrieval & Tool Use

Contact

Pinned Loading

  1. Charisma-Predictor Charisma-Predictor Public

    Multimodal AI pipeline to predict Big Five personality traits and assess charismatic leadership using audio, text, and video inputs.

    Jupyter Notebook 7

  2. MiniVision-Lightweight-and-Transformer-Models-for-CIFAR MiniVision-Lightweight-and-Transformer-Models-for-CIFAR Public

    Benchmarking CNNs and Vision Transformers on CIFAR-10/100 using a unified PyTorch pipeline with transfer learning and model fusion.

    Jupyter Notebook 1

  3. Image-Restoration Image-Restoration Public

    Deep learning-based image restoration pipeline with DnCNN, NAFNet, and legacy joint models. Includes PSNR/SSIM/LPIPS evaluation and visual comparisons.

    Jupyter Notebook 4

  4. Image-Stitching Image-Stitching Public

    Image stitching with Harris corner detection, SIFT descriptors, Lowe’s ratio test, and affine RANSAC warp.

    Jupyter Notebook 2