A short note.

I'm Tirop Meshack. I grew up in Kenya and went to Kaplelach High School. In 2022 I moved to South Korea to study Computer Engineering at Kyungdong University. I'm based in Goseong, in Gangwon-do, a quieter coastal area on the east coast. I'm now a senior, expecting to graduate in 2026.

My interest in computers started as the usual curiosity. Taking apart anything with a circuit board, breaking websites for fun, learning Python because the syntax made sense. The shape of what I do now took longer to come into focus. It sits at the intersection of three things: full-stack development (because I like shipping software people actually use), machine learning (because I like the modelling problem), and bioinformatics (because the data is hard and the stakes are real).

The work

Since 2022 I've been freelancing as a full-stack web and mobile developer, mostly Django and React Native. I'm currently the backend developer on Multition Education, a Kenyan AI-education platform serving 500+ students, and the web developer for Sunrise Drilling Limited. Tools I lean on for those: Django, PostgreSQL, Sentry for error monitoring, Grafana for observability, AWS for hosting. The longest-running personal project is JiraniSoko, a neighbourhood marketplace I'm building for Kenyan urban communities. Listings, jobs, motors, property, all wired up to M-Pesa.

The research side runs in parallel. My senior thesis, under Prof. Zubaer Ibna Mannan in Smart Computing at Kyungdong, was on temporal prediction of heart defects from chest X-rays with the CheXchoNet dataset, a CNN that reached AUC 0.79 to 0.84 in cross-validation. It pointed me towards waveform-level signals.

The current research is the next step in that direction: non-invasive intracranial pressure (ICP) estimation in pediatric neurocritical care. I'm training LSTM, TCN, and Transformer models on PhysioNet's pediatric waveform dataset, with PhysioNet's CHARIS as a cross-dataset validation set. The current best LSTM reaches a mean absolute error of 2.92 mmHg over 12 patients in leave-one-patient-out cross-validation. The full picture is on the research page.

Earlier work: a binary classifier on the Cleveland Clinic cardiovascular dataset (96% test accuracy). It was the project that first pulled me into clinical-data ML.

What I'm aiming at

I want to keep working at the intersection of ML and clinical signals, particularly cardiovascular and neurocritical care, where the data is rich and the failure modes are expensive. In the longer run I'd like to contribute to open-source research tooling: the kind of small, sharp libraries that make a researcher's day easier.

In the meantime, freelance work pays the bills and keeps me sharp on the parts of the stack that papers don't cover. Auth, deployment, payments, the boring shipped-and-monitored kind of software.

Outside of code

I read more than I write. I'm a member of IEEE and try to follow the bioinformatics track. My Korean is at the "buy groceries, miss jokes" stage; English is fluent and Swahili is home.