Hayden Housen

Co-founder and CTO @ Treblo (YC W24)

Treblo is an AI music editor that turns prompts, lyrics, or melodies into full songs in any style.

Software Engineering Intern @ Vocode (YC W23)

• Collaborated directly with founders as a pivotal engineer to shape Vocode’s pioneering AI-driven call automation solutions.
• Demonstrated versatility as a fullstack developer by building the FastAPI backend, open-source self-hosted service (Python and asyncio), and frontend dashboard (Next.js), while also resolving issues and reviewing pull requests from the community.
• Evaluated the latencies of multiple LLMs, speech-to-text APIs, and synthesis services to enhance Vocode’s performance.

Undergraduate Researcher @ Cornell

• As a sophomore, researched unsupervised object discovery and the scaling capabilities of reinforcement learning models (such as DreamerV3) with Dr. Ellis.
• As a freshman, overcame bias in paraphrase identification by using autoregressive transformers & out-of-distribution detection techniques: “GAPX: Generalized Autoregressive Paraphrase-Identification X.” Published in NeurIPS 2022 (3rd author). Advised by Dr. Sernam Lim at Meta AI. Also, experimented with zero-shot image classification via OpenAI’s CLIP model.

Machine Learning Intern @ Ada Support

• Improved Ada’s production accuracy by 8% using only 3% of production data by developing a novel intent classification pipeline.
• Conducted >60 experiments and trained >110 models to determine the most accurate methodology.
• Experimented with knowledge transfer, unsupervised learning of sentence embeddings, multi-task learning, and contrastive losses in the context of transformers and support vector machines.

Machine Learning Intern @ Ada Support

• Led the discovery and experimentation phases of a project to enable Ada chatbots to better understand non-English languages.
• Wrote a data processing pipeline to efficiently clean and analyze 9 billion chat messages for machine learning models.
• Researched novel techniques in multilingual intent prediction and cultivated skills in PyTorch, transformers, and pandas.

lecture2notes

• Created a state-of-the-art system to summarize classroom lectures using PyTorch, transformers (BERT), optical character recognition, speech to text, and convolutional neural networks. Source on GitHub. Learn more in the research paper.
• Named a top 300 scholar in the 2021 Regeneron Science Talent Search, the nation’s oldest and most prestigious science and math competition for high school seniors.
• Deployed ML pipeline in production via a full-stack website powered by Docker, Flask, Celery, Bootstrap, and Stripe.
• More info on the project page.

TransformerSum

• Furthered research in neural-network text summarization with a focus on long document summarization. 310+ stars on GitHub.
• 4.45x smaller than the state-of-the-art model but 94% as accurate at release. 10+ pre-trained models available.
• Rewrote researchers' code with enhanced performance and a focus on code readability and thorough documentation.
• More info on the project page.

Cybersecurity Challenges

• Placed 23rd (top 0.7%) in 2021 National Cyber Scholarship Competition.
• Placed in top 3% on average in PicoCTF 2019/2021/2022 competitions. Wrote technical guides with over 94,000 views to document my learning and help others.
• Continuously practicing cybersecurity principles by solving HackTheBox.com machines and publishing writeups.
• Learned ethical hacking skills including web exploitation, cryptography, reverse engineering, and binary exploitation. Worked with popular tools included in Kali Linux.

Cornell University

Info: College of Engineering. Dean’s Honor List. Previous president of the Cornell Cybersecurity Club.

Coursework: Machine Learning, Artificial Intelligence, Functional Programming, Linear Algebra, Probability and Statistics