Project 04

Teknofest E-Commerce Recommendation

Team lead for a learning-to-rank recommendation system on a Trendyol e-commerce dataset at the Teknofest hackathon — my first time owning other people's deadlines.

§ 01 Problem

Given real-world-scale e-commerce interaction data, rank the products each customer is most likely to engage with — under hackathon time pressure, with a team that needed the problem framed before anyone could start building.

§ 02 Approach

  • Led the team end to end: decomposed the problem into ranking target, feature pipeline, and evaluation, split the work along those seams, and owned the feature pipeline myself (feature.py).
  • Engineered user-product interaction features and ran Optuna-driven hyperparameter search over the ranking model (optuna.ipynb) — my first serious contact with structured search instead of hand-tuning.

§ 03 Outcome

A working end-to-end submission — and, more durably, the experience of being the person a team looks at when the plan has to change mid-hackathon. The leadership lesson outlasted the leaderboard position.

§ 04 Evidence

evidencetrendyol_ranking_projesi — feature.py, optuna.ipynb
Teknofest e-commerce recommendation submission
fig. 1 — Teknofest e-commerce recommendation submission