• Evaluating the neocortex’s ability as an autoencoder by feeding images to a neocortex model and training a machine learning model to reconstruct the images from real-time neuron activity• Investigated how the inhibition of excitatory neurons and neuron proximity influenced the neocortex’s latent representation of visual data

- Working with a team to create detailed lessons, assessments, and instructional materials for Multivariable Calculus that are accessible and engaging for students- Using PERL and LaTeX based language to write questions

- Owner, barber, and social media manager for Joyocuts barbershop in partnership with co-founder Miles Zhou.- Skills learned include Excel and Typescript for Excel, video editing, and social media management.- Currently have 300+ customers with almost 500+ haircuts done.

- Improved grounding through novel embedding methodology in web agents which use state of the art Large Language Models (LLMs) (e.g. GPT-4o and LLaMa)- Developed new method of embedding long context HTML incorporating DOM tree structure to improve grounding in the real world environment- Used novel methods (e.g. saliency matrices) to analyze weights and attention across different layers of LLMs to locate issues with previous embedding methods

- Created an AI copilot with biotech domain knowledge using a pre-trained LLM through Hugging Face- Used PyTorch and vector databases for knowledge embedding for enhanced performance.- Used RAG architecture to reduce hallucinations and improve accuracy on domain

- Worked as a food service employee with responsibilities such as customer service, employee training, etc.