About me

I'm Arel, a 16-year-old rising senior at Adlai Stevenson High School in Illinois. I've been into computers for about as long as I can remember — I taught myself to code, and then taught myself machine learning the same way, by building things until they worked. Everything on this site is a project I took end to end, from the dataset to the trained model to an app or a demo you can actually try.

Why medicine? That part is personal. When my 13-year-old cousin was diagnosed with leukemia, I couldn't wrap my head around how that happens to a kid, so I did the one thing I know how to do: I tried to understand it through code, by building a model that classifies leukemia cells. She beat it and is doing well now — but it stuck with me. It also sits right where I come from. We're a family of four: my dad is a computer entrepreneur and my mom works in medical labs and the biological sciences, so the place where computing and the life sciences meet has always felt like home.

The thing I care about most is being honest about what these models can and can't do. It's easy to post a 99.8% and move on; I'd rather show the patient-level result that dropped to 81%, because when you're modeling a disease a real kid you love actually had, an inflated number isn't a win — it's a lie. So every project here reports real metrics on data it hasn't seen and says plainly where it falls short. None of it is a medical device — it's how I'm teaching myself to build tools that could one day be trustworthy enough to actually help.

I'm working toward studying Computer Science + Bioengineering — the one path where I don't have to choose between the two things I love. More on that here →

About this portfolio

RLcapstone.ai presents a series of end-to-end applied AI projects, with a focus on biomedical computing. Each project is developed in full — dataset preparation, model training on local GPU hardware, rigorous evaluation, and deployment to a functional application or interactive demonstration — and is documented with an emphasis on reproducibility and the honest reporting of limitations.

The write-ups aim to document not only the results but the engineering decisions and pitfalls typically omitted from tutorials: platform-specific export toolchains, dependency constraints that silently break builds, thresholds chosen for principled reasons, and results that are strong but not without caveats.

Methodology

• Training — PyTorch on local NVIDIA GPU hardware, using public research datasets.
• Deployment — on-device inference wherever feasible (TensorFlow Lite, ONNX Runtime), so demonstrations run offline and no user data leaves the device.
• Applications — Flutter for mobile, static web pages for browser demonstrations.
• Evaluation — every project reports metrics on held-out data and states its limitations explicitly.

Source code

The full source code for these projects is available on GitHub: github.com/anotherhuman5345/rlcapstone. Questions and corrections are welcome there.