Projects
Yent
Yent: A Semi-Anonymous Social Media / Chat App
Yent is a semi-anonymous social media and chat app that offers unique privacy features. Group members can see who else is in the group, but posts are anonymous by default with rotating usernames. Users also have the ability to flag posts or members for removal in a democratic fashion, ensuring a community-driven experience.
The app currently features a Node.js backend with WebSockets for real-time communication. I’m in the process of migrating to AWS infrastructure, using API Gateway, Lambda, DynamoDB, and RDS to enhance scalability and performance. Additionally, the app will integrate AWS Cognito User Pools with Google login for user authentication.
This project also marks my first foray into mobile app development, using Expo and React Native for the mobile client. As a hobby and side project, Yent is still in active development, with many exciting features on the horizon.
Scratchy Grad
Handcrafted Autograd Engine with MNIST Visualization
I engineered a custom autograd engine in TypeScript, designed from the ground up to compute gradients and visualize backpropagation through computational graphs. To showcase its capabilities, I built an intuitive interface where users can interactively explore the flow of values and gradients. This engine also powers a dynamic MNIST digit recognition tool, complete with training visualization and performance metrics, bridging the gap between theory and application.
Smol language models (SLAMs)
In-Browser Language Models: Exploring NLP Through React
This project is a collection of lightweight language models that train directly in the browser, offering a hands-on introduction to natural language processing concepts. Built with React and TypeScript, these models demonstrate the progression from foundational techniques to more advanced architectures, showcasing both linguistic theory and software engineering principles. Highlights include:
- Foundational models like bigram and basic MLP implementations.
- Enhanced versions leveraging TensorFlow.js for improved training and performance.
- Upcoming projects exploring state-of-the-art techniques such as WaveNet and transformer architectures.
Angry Nerds
Physics-Powered Angry Birds-Inspired Game
I created a dynamic, physics-driven game inspired by Angry Birds, built entirely with JavaScript and the HTML5 Canvas API. At its core is a custom physics engine I developed, enabling realistic gravity, friction, and collision detection. This project combines creative gameplay mechanics with technical problem-solving to deliver a fun and interactive experience.