This is the version of me that doesn’t fit on a PDF: the things I’ve built, the side projects I’m proud of, and a few demos you can actually play with. Agents did a lot of the building; I did the editing.
Welcome,
Relevant experience
MLE + SWE · BEng Computer Science, University of Edinburgh, 2026
I’m the person who can train the model and ship the thing people use. Sometimes, debugging a PyTorch shape error and other times steering a coding agent through a build.
On Nethermind’s hardware ML team I pitched an idea, generating synthetic palm-vein images with a diffusion model, and then built a pipeline for it, from pre-processing to inference. I had a lot of fun there and worked on plenty more. A happy moment: better dataset filtering moved a business metric in my first week.
I collaborated with Prof. Jeremy Bejarano on a forecasting benchmark and wrote the foundational ML code in the repo: the first OOP APIs, which put three forecasting libraries behind one interface so training or inference takes a few lines. I also made it work on Lambda AI GPUs, created the workflow automations and the pandas pre-processing. The project became an OFR working paper.
What happens when deep models get anywhere near nuclear command and control? I spent a fellowship digging into that and wrote a 14-page paper on where AI could be used in NC3 and the technical challenges that come with it circling around explainability, robustness and safety. It was accepted to the 2025 SPSA summer conference.
Claude coded a hybrid search over arXiv and OpenAlex papers by conceptual similarity to a query. I added a “core idea” per paper, which retrieved better than abstract embeddings.
Hybrid search over arXiv and OpenAlex papers
Type an idea and find papers that are about the same thing, even when they use different words. Each paper gets a generated “core idea”, which made search 21% better than using abstract embeddings. It all runs on Cloudflare.
Sketch-RNN doodles, drawn stroke by stroke
Pick two numbers and watch a little neural network doodle for you, stroke by stroke, in real time.
Your two numbers are coordinates: the spot they land on picks one of twenty Quick, Draw! classes. The model, Sketch-RNN, runs entirely in your browser.
I’ve been helping my dad with his business mostly on the digital side. It started with Canva designs and Meta ads. Before agents existed I was building WordPress sites in Elementor, then Shopify stores, hand-made from templates. This year I built him three websites with agentic development and set up SEO/GEO Claude skills.
It was never only building. I also sat in on decisions and planning, and picked up skills on the go. That’s why AI feels natural to me: stacks change, but the principles and the decisions behind them mostly stay the same.
Languages show it. We now lean on type-safe languages, and my C++ and Java make static typing comfortable. I’ve also used Python heavily for ML, so I know what duck typing makes possible. The same goes for OOP and other principles: knowing when they help is what lets me steer coding agents.
Qwen3.5-2B fine-tuned with LoRA on CUAD
Upload a contract, see the risky clauses highlighted.
Under the hood, Qwen3.5-2B is fine-tuned with LoRA on CUAD, served on Modal behind a Cloudflare Worker.
Advanced Computing for Finance, UChicago (graduate course)
My team and I implemented an HFT prototype in C++, with an orderbook, a matching engine and other components.
We used Robin Hood hashing, priority queues, cache locality and benchmarking analysis, and custom data structures to store and retrieve orders and market data. We also used SIMD programming for basic time-series analysis with moving averages.
Full-stack Quantitative Finance, UChicago (graduate course)
My project partner and I recreated the Treasury swaps data and plot from the paper Segmented Arbitrage (Siriwardane et al.).
The project is automated and reproducible: unit tests run with Pytest, Pydoit automates the tasks, and the LaTeX report is generated automatically.
I majored in Computer Science, and, at university, went deeper and deeper into ML. Together, the two gave me a feel for the whole AI stack, from the data to the product.
Nethermind · Edinburgh
Hardware division ML team, working on palm-vein biometric verification. Synthetic data with diffusion models, ViT training and evaluation, GPU-offloaded embedding search, and GAN-based pen testing.
Prof. Jeremy Bejarano (UChicago)
Financial time-series forecasting benchmark. I wrote the foundational ML code in the repo: the initial OOP model APIs, GPU and workflow tests, and the data pre-processing pipeline. The project became an OFR working paper.
University of Chicago
Wrote a 14-page paper on AI in nuclear command, control and communications, accepted to the 2025 SPSA summer conference. Also an XLab AI Safety Intro Fellow and later an XLab member.
C-DAC
Proof-of-concept deep-learning password tool using a GAN and a Transformer, with a PyQt interface and GPU support.
The University of Edinburgh
Four years of CS, from computer architecture to distributed systems and cloud programming, plus a few graduate courses along the way. Graduated with a First.
University of Chicago
A year in Chicago, where I took graduate courses (including Fundamentals of Deep Learning at TTIC), got involved with XLab, and won the course HFT competition.
For the research collaboration with Prof. Jeremy Bejarano
It is awarded for completing tasks, spread out over three months, related to reflection and learnings from the work experience.
The tasks asked me to step back from the forecasting benchmark itself and write down what I had learned from it, and how I worked.
Stanford · Andrew Ng
Learned various ML concepts from Andrew Ng’s course. It set the foundations for my later learnings in DL.
Things like gradient descent, regularisation and the bias–variance trade-off came from here first, before graduate deep learning at TTIC built on some of them.
Let's chat about ML, building products, or anything on this page. Email is one of the quickest ways to reach me.
Prefer to ask your own AI?
Copy a prompt that points your assistant at this site, then ask it anything about me.
You made it to the bottom. Thank you, genuinely. I built this site to show more than a PDF can, and it means a lot that you scrolled this far.
If something made you smile, or something felt off, let me know! The note is anonymous and lands in my inbox, as long as nothing errors along the way.