Mathematics and Computer Science student at École Polytechnique, interested in applied machine learning and backend systems.

Most recently interned at British American Tobacco as a Data Analyst, built order-processing automation for the Trade team

Building an ML pipeline that predicts molecular mutagenicity from chemical structure, using scaffold-based splitting to test generalisation to unseen scaffolds

LinkedIn · kochenkoalesya@gmail.com

neural-network-activation-study — How activation function choice affects backpropagation and convergence. Backprop derived by hand, implemented in NumPy, and validated against the hand-computed gradients. Python NumPy Matplotlib

vessel-power-regression — Predicting ship power consumption from derived speed through water. Vector decomposition to recover speed through water from speed over ground and sea currents, then linear vs. polynomial regression. Python Pandas scikit-learn

python-auth-system — Flask authentication with bcrypt hashing, login rate limiting to blunt brute-force attempts, and email account activation using cryptographically secure tokens. Python Flask MySQL bcrypt

gastrogo — Full-stack food delivery platform built without a framework: hand-rolled routing, sessions, and a normalised ten-table schema with a self-referencing category tree. PHP MySQL PDO

Python · C/C++ · Java · SQL · PHP · Pandas · NumPy · scikit-learn · TensorFlow · MySQL