Computer Science + Bioengineering
This portfolio sits at the intersection of computer science and bioengineering. Each project pairs a concrete computational method — deep learning and biomedical signal or image processing — with a health-technology problem, and each is carried through the full lifecycle: dataset, model, rigorous evaluation, and a working application. The sections below map that work to the two disciplines.
Why I'm pursuing this major
I started my leukemia project for a specific reason. My 13-year-old cousin was diagnosed with leukemia, and I couldn't wrap my head around how that happens to a kid. I've always been a computer person — coding is how I make sense of things — so I did the one thing I knew how to do: I tried to understand her diagnosis through code, building a model that classifies leukemia cells from blood-smear images. She has since beaten it, and is working toward a normal life.
That project is also why I care so much about being honest about results. My first version scored 99.8%, and instead of celebrating I got suspicious — the dataset had no patient IDs, so the model was probably winning on the wrong cues. When the disease you're modeling is one a real kid you love actually had, a fake-high number isn't a win; it's a lie. So I rebuilt the experiment properly — a dataset with patient labels, tested only on patients the model had never seen — and the honest accuracy came out around 81%, not 99.8%. That lower, real number means far more to me than the inflated one.
CS + Bioengineering feels like home in an almost literal way. My dad works in computers and my mom is in the biological sciences, and I grew up loving both — this is the one path where I don't have to choose. It's also exactly what the CS + Bioengineering degree at Illinois is built around: computation applied to screening, diagnostics, and treatment. My EEG and ECG work lines up with Grainger's neural-engineering program (the first neural-engineering bachelor's in the country), and my imaging classifiers fit its bioimaging and computational-biology research.
What I'm still missing is the biology and signal-processing foundation to make my models rest on real understanding rather than just labeled data — and that's what I want to learn at Illinois, so I can build diagnostic tools trustworthy enough to actually help someone like my cousin.
How the portfolio maps to the two disciplines
The computer science throughline is applied machine learning done carefully: convolutional neural networks, leakage-aware evaluation, and the engineering needed to ship a model to a phone or a browser. The bioengineering throughline is the domain grounding: physiological signals (ECG, EEG), medical imaging (dermatology, hematology), and a clinical evaluation mindset that favors sensitivity, specificity, and ROC-AUC over raw accuracy. The recurring theme across both is responsible, honest health technology — privacy-preserving inference, clearly stated limitations, and explicitly non-diagnostic framing.
Project by project
MoleCheck
CS: transfer-learned CNN (YOLO11), model export to TensorFlow Lite, and
on-device mobile engineering in Flutter.
BIOE: dermatological image analysis for skin-cancer screening, tuned for
sensitivity and built to preserve the privacy of sensitive medical imagery.
ECG Arrhythmia Classification
CS: 1D convolutional network, signal-window preprocessing, and a
leakage-aware inter-patient evaluation protocol.
BIOE: cardiac electrophysiology — classifying heartbeats and confronting
the inter-patient generalization gap central to real ECG diagnostics.
ADHD Classification from EEG
CS: multi-channel 1D CNN with a strict subject-level split and
subject-averaged evaluation.
BIOE: neural signal analysis on 19-channel EEG, applying research
methodology that separates a genuine group signal from individual memorization.
Leukemia Cell Subtyping
CS: transfer-learned image classifier and a critical analysis of
dataset leakage behind an apparently perfect score.
BIOE: hematology and blood-smear microscopy — distinguishing leukemic
blast stages from benign look-alikes, with clinical validation caveats.
Skills across the two disciplines
Computer Science
- Deep learning: 1D and 2D convolutional neural networks across signals and images
- Rigorous ML methodology: leakage-aware and subject-level train/test splits
- Model deployment: PyTorch → ONNX / TensorFlow Lite, on-device and in-browser inference
- Full-stack delivery: Flutter mobile apps, static web demos, automated PDF reporting
- Reproducibility and verification: confirmed model-export parity and honest, documented reporting
Bioengineering
- Physiological signal processing: ECG (cardiac) and EEG (neural)
- Medical imaging: dermatoscopy and blood-smear microscopy
- Clinical evaluation: sensitivity, specificity, ROC-AUC, and macro-recall over accuracy
- Responsible health AI: privacy-preserving inference and honest, non-diagnostic framing
- Domain grounding: pathology look-alikes, inter-patient variability, and dataset equity