Ryan Heise

I developed an automated deep-learning video editing pipeline that generated over 100,000 YouTube views by combining Whisper for transcription, ChatGPT for highlight selection, S3FD for face tracking, TalkNet for active-speaker detection, and FFmpeg for rendering.

I built a compiler that translates a subset of Java into optimized 32-bit ARM assembly using Lex and Yacc for abstract syntax tree construction, and I abstracted parsing operations into modular abstract classes to ensure efficient, maintainable code generation.

I built a full-stack web application using Django, Django REST Framework, and React to manage my family’s annual football league. I deployed it on a Raspberry Pi with Gunicorn behind Nginx and set up custom network configurations to provide secure, reliable access.

I designed a modular election system that leverages homomorphic encryption for secure, local vote tallying and incorporates digital IDs alongside asymmetric cryptography to ensure end-to-end verifiability, protect voter privacy, and scale efficiently to large electorates.

I developed an active learning framework that evaluates various sampling and querying strategies across a wide range of models and datasets, and I then compiled the results into a survey-style report comparing performance metrics and recommending the most effective approaches.

I engineered a high-performance HTTP server in C++ using raw system calls and socket programming to serve HTML and image files. To maximize concurrency and throughput, I implemented both multiprocessing and multithreading architectures, allowing the server to handle multiple requests efficiently.

I created a web app with a small team that uses Bluetooth data to estimate restaurant wait times. It was really scuffed since I built it as a freshman, but somehow it won first place out of 55 teams at the Purdue Hello World hackathon.

I built a neural style transfer model from scratch using PyTorch that applies the artistic style of one image to the content of another. I implemented the algorithm by optimizing a loss function that combines content and style representations extracted from a pre-trained convolutional neural network.

I trained a gradient boosted decision tree to predict whether or not an individual was likely to default on a loan. This was for a Purdue competition and I placed 2nd out of 93 teams.

I built a convolutional neural network from scratch using PyTorch to classify images from the CIFAR-10 dataset. I implemented data augmentation techniques and experimented with different architectures to improve accuracy.

I implemented a Unix-like command line shell in C that supports features like piping, redirection, and job control. I used system calls to manage processes and handle user input, creating a functional and interactive shell environment.

I implemented a custom memory allocator in C that mimics the behavior of malloc and free. I used techniques like segregated free lists and boundary tags to manage memory efficiently and reduce fragmentation.

I constructed a full stack web application that allows me to track my progress on various goals using a video game like interface. I built the backend with Django, and the frontend with ReactJS.

I built a decentralized web application that lets users mint nfts of various purdue professors on the etherum blockchain. I built the smart contract with solidity and the frontend with react.

I built a web application that identifies the license plate of a vehicle and then uses that information to determine vehicle's the carbon emissions.

I built a light weight stenography program that would hide data inside the lsb of each pixel. I then integrated this into a small python website using the CPython api. The image above contains the Darth Plagueis the wise quote.