I joined MaltedAI in June, 2025 as a software engineer working on core AI.

• I joined HackerNoon as an intern during my penultimate year of high school and stayed on as a permanent developer throughout my first 3 years of university.
• HackerNoon 3.0: Lead the development of a full migration to SvelteKit & Postgres. Resulted in web vitals gains of over 20% and hosting cost reductions of ~40%.
• AI Infrastructure: Designed and implemented a scalable hosting infrastructure for internal models using Kubernetes and Google Cloud which powers the AI features in the editor.
• Metrics: Developed a distributed internal analytics system to accurately measure key metrics while maintaining user privacy.
• Editor: Improved the editor experience by implementing a variety of features including a custom markdown parser, real-time collaboration and a custom image editor.
• NFTs: Minted 20,000+ NFT awards for top authors using a custom smart contract on the Polygon blockchain.
• Comments & Notifications: Built a robust comment and notification system, including real-time updates, admin moderation panel and comment threading.

• Collaborated with the Edinburgh team to enhance FPGA testing processes.
• Learned VHDL and Verilog and wrote tests to detect defects in manufacturing.

• Designed and implemented a variety of deep learning models to detect the use of engines in online chess games.
• Generated a novel dataset using Maia & Stockfish to represent selective cheating.
• My model reached an accuracy of 92%, surpassing the Stanford PhD benchmark by ~14%.

• Built a deep learning model to generate guitar tabs from audio input using a custom dataset of 10,000+ songs.
• Utilised a transformer model with Mel-Spectogram encoding to learn the relationship between audio and tablature.
• Achieved an accuracy of 85% on a held-out test set, outperforming the state-of-the-art by 10%.

• Developed a custom autograd engine in Python to learn the fundamentals of backpropagation and gradient descent.
• Implemented a variety of optimisers including SGD, Adam and RMSProp to demonstrate the differences in convergence speed.
• Used the engine to train a simple neural network on the MNIST dataset to achieve 95% accuracy.

• Built a web app using SvelteKit, Turso & CloudFlare to help people find the best pubs and bars in Scotland based on community reviews.
• Utilised the MapBox API to display nearby pubs and bars and a custom review system to allow users to rate and review locations.
• Achieved over 500 active users and over 300 reviews before being acquired.

• Built a robot capable of autonomously navigating a maze towards a user using SLAM & frontier exploration.
• Utilised a ViT image classification model to detect the type of waste.
• Manages multiple users and locations using a FIFO queue.
• Won the SkyScanner Most Sustainable Project award.