Projects

Co-developed Spy-Search, a modular multimodal research assistant for retrieval and question answering, integrating LangChain, LLMs, and multimodal document understanding. The project has received 385+ GitHub stars and 48+ forks.

A spatiotemporal deep learning framework for needle-tip localization in ultrasound videos using both spatial and temporal features.

Refines the hidden states of a frozen vision-language model in latent space with an RL-trained controller before decoding, improving multimodal math reasoning without generating chain-of-thought tokens.

Multi-role debate simulation with GPT. Users define a topic, and LLMs autonomously engage in multi-perspective discussions.

A technical survey on watermarking and traceability methods for diffusion-based generative models.

A real-time food ordering platform built with Go and Firebase.

An interactive chess game implemented in C++ with ImGui, including sound effects and real-time statistics.

Core machine learning methods, covering clustering and outlier detection, regression with feature scaling, and learning-curve analysis.

Language modeling from classical to modern architectures: n-gram, RNN and LSTM language identification, self-attention for semantic similarity, gated state-space models for key–value memory, and a masked diffusion language model trained on TinyStories.

Deep learning for medical data: pneumonia classification from chest X-rays, EEG motor-imagery decoding with EEGNet and DeepConvNet, and multi-class chest X-ray diagnosis (Normal / Bacteria / Virus / COVID-19).