I CONSTRUCT CONNECTIONS BETWEEN EMERGING TECHNOLOGIES AND HUMAN
I am a passionate software engineer who builds with both purpose and spirit, blending technology and curiosity to explore deeper connections between AI, software, and human experience.
I am a passionate software engineer who builds with both purpose and spirit, blending technology and curiosity to explore deeper connections between AI, software, and human experience.
SELECTED PROJECTS
Hover description to preview projects
Engineered a 53M-parameter GPT-2 architecture from scratch using Apple’s MLX framework. Implemented Pre-LayerNorm transformer with 8 layers, achieving 27.5K tokens/s training throughput and 169 tokens/s inference on Apple M2 Pro. Trained on 10M tokens of TinyStories dataset for 20K iterations (loss: 0.758). Model available on Hugging Face.
Developed a cognitive humanoid navigation framework combining LLM planning with Unitree RL policy controller for H1 robot. Architected 4-layer system integrating natural language understanding, C++ optimized perception (pybind11, less 10ms LIDAR processing), A* path planning, and adaptive motion control. Solved critical navigation challenges: speed parameter propagation, heading error correction, and autonomous stopping (1.2m tolerance). Achieved stable navigation at configurable speeds (0.5-3.0 m/s) in MuJoCo simulation.
JasOS-Kernel is a Unix-inspired, bare-metal operating system kernel built from scratch as a personal project, designed to run on ARM architecture using the QEMU emulator. It showcases core OS concepts such as memory management and process scheduling
developed a specialized AI assistant by integrating the GROQ API, the LangChain framework and the Chroma vector database to enhance the interpretation capabilities of files. Optimized the processing and retrieval of PDF materials, enabling efficient and interactive learning experiences. This solution improved accuracy and response speed, streamlining how users engage with academic content.
Developing a Generative Pre-trained Transformer (GPT) based on the architecture outlined in the paper 'Attention is All You Need.' This involves understanding and implementing the underlying mechanisms, including activation functions, loss functions, and normalization techniques, to gain a deeper insight into the model's operation and performance.
This software leverages OCR technology to search for questions and answers in past exam papers. Currently, it’s in the prototype phase, with ongoing improvements in development. While the system is functional, it indexes each page individually, which impacts its efficiency.
This project involves developing a neural network classifier to predict lottery numbers. The model explores patterns in historical data, aiming to improve prediction accuracy while leveraging deep learning techniques. Currently, it’s an experimental approach focused on optimizing model performance.
A dynamic street-fighting game where players face off against opponents in gritty, urban environments. The project focuses on fast-paced combat mechanics, unique character abilities, and engaging visuals. The game aims to deliver an exciting arcade-style experience.
Papering is a Python-based tool designed to help students quickly find A-Level exam papers. Using Beautiful Soup for web scraping and WebBrowser for easy access, it streamlines the process of searching and retrieving relevant papers. The interface, built with PyQt5, ensures a user-friendly experience.
Click "more" button or the "GitHub icon" located in front of the Project title to access my GitHub. Hope you finding something inspiring and exciting!