Hello!
I'm Caleb Penley,
a computer science student with a deep interest in researching and designing AI systems inspired by human cognition.
Get in touch: penleyc@icloud.com
Currently attending The Pennsylvania State University, pursuing a degree in computer science, I have a keen interest in the intersection between machine learning and neuroscience. I'm currently focused on experimenting with system and architectural design that more closely mimic the efficient adaptability seen in biological neural networks.
Growing up in a small rural town in northwestern PA, I was surprisingly drawn to the complexity of computation, eventually reverse-engineering everything from engines to computer motherboards. While expanding my knowledge through experimentation, I have discovered I am fascinated by a variety of disciplines.
When I’m not immersed in theory or coding a custom tool, I enjoy reflecting on life's big questions, going on hikes, and socializing with friends and family.
- Python
- C
- Rust
- Java
- Tech Ops
- Machine Learning
- Digital Systems
- Automation
- Teamwork
- Communication
- Coordination
- Mentoring
A full-stack asynchronous dashboard with a TCP backend for live projector status and management. Secured by UUID-based authentication and automated email alerts for critical errors or projector failures.
Original AI assistant with a Flask REST API for CI/CD, media playback, and real-time weather updates. Scaled to support a server-client design using multi-threaded sockets, leveraging Microsoft Azure for speech recognition and synthesis.
A personal research project exploring fine-tuning of a 1B-parameter Llama model to perform textual style transfer using a limited dataset. Examined the effects that supervised fine-tuning (SFT) and input/output variations have on model quality and accuracy.
A mdadm-like tool in C for RAID management, focusing on fail-safe Just a Bunch of Disks (JBOD) operations through rigorous testing. Implemented a caching system to optimize I/O and socket-based communication for scalability and improved performance in a networked environment.
A convolutional VAE in PyTorch from scratch to efficiently learn compressed representations from images. Trained on the CIFAR-10 dataset, optimizing for low reconstruction loss while achieving diverse generative sampling.
A personal project exploring the capacity of a lightweight LLM like TinyLlama to perform structured textual reasoning through deductive prediction.
A Java-based academic scheduling system using SQL, supporting dynamic student course enrollment and admin-level schedule control.
Implemented and trained Low-Rank Adaptation modules on a Flux.1 Dev to experiment with efficient model fine-tuning and performance improvements on domain-specific input.
A dynamic movie listings web app using TMDB API for movie data, backed by Firebase and Google Cloud Platform (GCP) functions for reliable showtime access.
Source Code
A simple yet clean and intuitive website designed to showcase pricing, events, and a gallery for my dad’s local paintball business.
Source Code
