Will Stephen
About
I’m a final-year Computer Science student at UCL with a background in audio engineering. I’m minoring in Intelligent Systems, UCL’s machine learning pathway. This summer I was a technology intern at Vanguard, where I built an AI agent within an existing Java backend. I like working on agents, backend systems, low-level code and machine learning.
Projects
What it is Built with Demo- MIDI Synthesiser2025–26 A polyphonic synthesiser with a hand-written Java engine, playable in the browser with a sequencer built for this site. Java, JavaFX, Rust, WebAssembly
- Manuscripta2025–26Coursework A closed loop for the classroom, built on local LLMs: personalise and produce lesson materials, deliver them to e-ink tablets, and mark with assistive AI. Android, Java, Electron, React, .NET, TCP/UDP
- research-loop2026 An autonomous research agent that explores the hypothesis space of a verifiable problem. Shell, Markdown skill files, ndjson, cron/launchd
- Neural Audio Quantiser2026 A neural approach to correcting the timing of recorded musical performances. Python, NumPy, PyTorch
- Braccio Controller2024Coursework An early foray into robotics and sensors: an Arduino arm that finds, picks up and colour-sorts blocks. C++, Arduino, Blender
- Self-balancing search trees2025Coursework Three self-balancing search trees benchmarked in Python, and how they differ on the same keys. Python, Jupyter, Matplotlib
Work
I spent the summer on the Funds of Funds team, within Vanguard's Global Investment & Financial Services (GIFS) division, building a proof-of-concept AI agent inside a large Java application that manages a multi-trillion-dollar asset base.
Questions about how the application behaved, or why an error had happened, used to take hours to piece together. The agent answers them by reading the documentation, the code, the logs and the database. I built its harness, agent loop and tools in the Java backend, with AWS Bedrock for the model and for retrieval.
I followed the codebase's domain-driven design, tested with JUnit and Playwright, and reviewed everything the coding agents wrote. Along the way I found and fixed a race condition in production.
Sole operator of an independent audio production business. Full technical and creative ownership across recording, mixing, and mastering for commercial clients. Built and maintained a custom studio environment, managing hardware integration and software toolchain configuration. Led multi-person recording sessions, coordinating musicians and managing project delivery to client deadlines.
Fulfilled customer orders for a premium gift brand, ensuring timely and accurate customisation and shipment while maintaining high quality standards.
Proactively identified process improvement opportunities and designed custom packing stations using 3D modelling, laser cutting, and assembly skills to optimise workflow.
Recommendation from Head of People highlighted contributions driving "both immediate results and long-term improvements."
Represented a cohort of 20 foundation-year students.
Surveyed and collected structured feedback on course delivery, built strong individual relationships with each student, and communicated findings to department staff to drive improvements.
Education
Minor in Intelligent Systems. This year: Machine Learning and Neural Computing, Machine Learning for Visual Computing.