About me
I am a Year 13 kid at Harrow School. My long-term goal is to advance explainable systems that possess human-like intelligence.
I have previously:
- Developed multimodal language models to interface with scientific data, under the mentorship of Dr Siddharth Mishra-Sharma and Prof Jesse Thaler at IAIFI, MIT (as part of Research Science Institute).
- Led research on the large-scale graph structure of Wikipedia links, culminating in a first author publication.
- Investigated using autoencoders for low-resource machine translation under Dr Alifierakis at Global Talent Mentoring.
Outside of research, I like making stuff because it’s fun. I started Team Enigma, leading the development of an AI-powered acoustic monitoring device for environmental conservation and prototyping a facial recognition system for registration in my boarding school. In NLP, I’ve experimented with automatic Haiku and traditional Chinese poetry generation, as well as finetuning GPT-2 to write for my school’s newspaper.
I also enjoy doodling with programming languages and built a pseudocode-to-Python transpiler completely from scratch, making it actually possible to run the pseudocode in my school curriculum. I also developed a novel tool to simulate a C++ interpreter on top of an existing GCC installation.
Currently, I’m interested in building models that align astronomical observations across modalities and the neural circuitry of small, multimodal models trained on synthetic test datasets.
At university and beyond, I hope to: 1) build a sound understanding of how and why ML models work in controllable contexts, using tools from physics, cognitive sciences, mechanistic interpretability, or elsewhere; 2) use this knowledge to develop systems with explicit reasoning, better interpretability, and less supervision.
My other interests include learning ancient languages, creative writing, photography, and philosophy. I have experimented with constructing my own language. I enjoy climbing, camping, and hiking.