About

I am Lethabo (Scofield), an AI/ML Engineer specializing in neural networks and machine learning systems. My work focuses on understanding how models perceive and represent information, reason across inputs, and adapt to complex environments, including reinforcement learning (RL) techniques such as those described in Proximal Policy Optimization (PPO).

My experiments are guided by state-of-the-art research. For example, in representation learning, I reference Transformers for attention-based models. In vision-language alignment, I study ALBEF: Align Before Fuse and CLIP: Contrastive Language-Image Pre-training. Each experiment integrates these insights to test model reasoning and semantic understanding across modalities.

I run Olyxee, a project dedicated to deploying neural networks on heterogeneous edge devices. The approach takes inspiration from research on model quantization and optimization for limited-resource environments, such as Deep Compression.

This site documents experiments, research notes, and small projects. Each post links directly to the research foundations behind the work, showing how theoretical insights are applied in practical neural network design and multimodal reasoning.