Profile

Turns raw data into clear answers that people can check. MSc in Artificial Intelligence (Machine Learning) from Liverpool John Moores University (taught-module average 78.8%; 83% in Advanced Topics in Deep Learning) and a BSc in Software Engineering. Looking for a graduate data, machine learning or AI role; based in Liverpool, open to Manchester and London.

​

Education

MSc Artificial Intelligence (Machine Learning), Sep. 2025 – Sep. 2026
Liverpool John Moores University, Liverpool, UK. Award pending; taught-module average 78.8%.

BSc Software Engineering (taught in English), Sep. 2020 – Jun. 2024
Yaşar University, İzmir, Turkey.

​

MSc Dissertation: Can AI Catch Phishing Emails Written by AI?

• Data: Built three clean datasets of genuine and phishing emails from different eras (29,224, 51,839 and 19,164 emails, the last generated by Google’s Gemini), removing duplicates and overlaps so results could not be inflated. Tools: Python, pandas, NumPy.

• Fair comparison: Compared simple models against large modern AI models (BERT, RoBERTa), with the rules fixed before seeing any results. Tools: scikit-learn, PyTorch, Hugging Face, Optuna.

• Result: The simple model caught AI-written phishing more reliably than RoBERTa, with fewer false alarms (8.3 vs 14.7 per 1,000 genuine emails), a statistically significant difference. The real weakness: models trained on older emails miss many AI-written ones (ROC-AUC fell from about 0.999 to 0.8429).

• Explaining decisions: Showed which words drove each decision, tested whether the explanations could be trusted, and found and fixed a hidden error in one method. Tools: SHAP, LIME, Integrated Gradients.

​

Other MSc Projects

• Question answering for NHS staff training (Advanced Topics in Deep Learning, 83%): Built 5,056 question-answer pairs from NICE and international clinical guidelines and fine-tuned four BERT models. The best (PubMedBERT) reached F1 76.67 vs 56.45 for a ready-made model, with calibration and safety-threshold checks. Tools: PyTorch, Hugging Face, FAISS.

• Wildlife object detection (Deep Learning Concepts and Techniques): Trained and compared two Faster R-CNN detectors on about 3,000 camera-trap images on a GPU, measured them with mAP at IoU 0.5 and 0.75, and studied where they failed (small or hidden animals, over-confident scores). Tools: TensorFlow Object Detection API, ResNet-50, TensorBoard.

 

Experience

Co-Founder and Technical Lead, Apr. 2026 – Present
FitEscala: helps independent gyms in Spain win new members, with a growth guarantee in the contract. Barcelona (Remote).

• Built the company website and a Python tool that finds and qualifies potential gym clients in Spain’s five largest cities. Tools: Python, HTML, CSS.

• Built the customer management system (CRM) the business runs on. It records every call, visit and sign-up for each gym, so both sides can check whether the growth guarantee was met. Tools: Node.js, Express, SQLite, Railway.

AI Trainer (Coding), then Coding Expert, Mar. 2025 – Aug. 2025
Outlier (Scale AI): contract work on a platform that trains AI models. Remote.

• Wrote, reviewed and rated code answers that are used to train and test large language models. Promoted from AI Trainer to Coding Expert.

Web Developer, Jan. 2024 – Mar. 2025
Mavi Pencere Special Education Association (EU-funded project), İzmir, Turkey.

• Built the website of an EU-funded project that helps people who use sign language communicate in healthcare, front end and back end. Made it accessible and fast: pages start to appear in under a second, even on low-end phones. Tools: HTML, CSS, JavaScript, PHP.

Software Developer Intern, Jul. 2023 – Sep. 2023
Malton London, London (Remote).

​

Technical Skills

Machine learning: scikit-learn, XGBoost, LightGBM, PyTorch, TensorFlow, Hugging Face Transformers, Optuna
Data analysis: Python, pandas, NumPy, Jupyter, SQL, PostgreSQL, RAPIDS; bootstrap tests, McNemar
AI engineering: OpenAI, Claude and Gemini APIs, embeddings and similarity search (pgvector), Docker, Git
Spoken languages: Turkish (native), English (C2)

​