Taha Berk Terekli

Data engineering & applied research

built to still work when you run it twice.

Now

Data intern at Getir, in Istanbul. SQL and Python over Amazon Redshift, pipelines scheduled with Airflow on AWS, and the dashboards in Tableau and Metabase that people actually open on Monday morning.

On the side I hold a TÜBİTAK 2209-A grant to study evaluation methodology for medical-image segmentation. Mostly it is an argument about what a number is allowed to claim.

Projects

Online Dershanem

An exam-prep platform in daily production use — lessons, mock exams, CRM and reporting behind one account system.

BraTS U-Net pipeline

A leakage-safe evaluation pipeline for brain-tumor segmentation. Frozen split, one guarded test run.

OneRead

One human-reviewed article briefing every weekday, with verified sign-up and billing.

OneGames

Three daily puzzles — Sudoku, DNA and Word — sharing one product shell with OneRead.

Quantum Groups

Symbolic Python verification of U_q(sl₂) and GL_q(2|1) — Hopf axioms and Yang–Baxter, checked by test.

Longer write-ups live on the projects page.

Previously

Before Getir I built RPA at Papara — Automation Anywhere workflows across .NET, VB and SQL — and before that at Latro, automating SAP processes with Robusta, Python and Java. In between, a supply-chain stint at LC Waikiki doing demand planning and inventory analytics.

Two years of automation taught me to distrust anything that "usually works." A bot doesn't get to shrug at an edge case. The full history is on the work page.

Writing

Why I'd rather build the same Redshift query twice

A defence of boring, idempotent ELT — and what it costs to enforce.

From RPA to data: notes from the in-between

What two years of automation work taught me before I moved into a data role.

What U-Net taught me about reading MRI slices

Field notes from the TÜBİTAK project — normalisation, patch sampling, honest evaluation.

all notes

Stack

Python, SQL, TypeScript. Redshift, PostgreSQL, Airflow, dbt-shaped ELT. TensorFlow and Keras for the research side; Next.js, Prisma and Docker for the things that ship.

Mostly pipelines, evaluation, and making results reproducible.

Studying

B.Sc. Mathematics at Yıldız Technical University (2022–2026). I came to data work from proofs rather than a CS track, and it shows — my first question about a system is what the definitions are, and what happens when the world stops matching them.

Elsewhere

Taha Berk Terekli 👋

Taha Berk Terekli