Hey,
I am Zeynep Tandogan

About

Passionate computer scientist and EPFL MSc student specializing in data science, machine learning, and large language models (LLMs).

Experienced in building scalable AI systems, from LLM alignment to Mixture-of-Experts optimization with Megatron-LM.

Research experience spans LLM alignment, scalable MoE optimization, and AI-powered document processing, including segmentation pipelines and LSTM/Transformer-based text transcription with beam search.

Skilled in the full ML pipeline, including multi-GPU training, model evaluation, and deployment.

Motivated by the challenge of creating scalable, aligned AI systems that serve human needs.

Skills

Education

• MSc in Computer Science @ EPFL (École Polytechnique Fédérale de Lausanne)

  Sept 2023 – March 2026

  Lausanne, Switzerland

  • CGPA: 5.25/6.00
  • Specialization: Data Analytics
  • Core Modules: Machine Learning, Modern Natural Language Processing, Applied Data Analysis, Reinforcement Learning

• BSc in Computer Science and Engineering & Electronics Engineering (Double Major) @ Sabancı University

  Sept 2017 – June 2023

  Istanbul, Turkey

  • CGPA: 3.96/4.00
  • Minor: Business Analytics
  • Distinctive student with full 100% merit scholarship

Experience

• Master Thesis Project Intern @ Swisscom

  Sept 2025 – Feb 2026

  Lausanne, Switzerland

  • Planned thesis on optimizing LLMs for segmented, multi-domain business data.
  • Explored segment-aware fine-tuning, modular input pipelines, and prompt engineering to adapt models across domains.

• Student Researcher @ Machine Learning and Optimization Lab at EPFL

  Feb 2025 – July 2025

  Lausanne, Switzerland

  • Developed optimization strategies to improve scalability and efficiency in Mixture-of-Experts (MoE) models.
  • Implemented a custom MoE architecture and ran multi-GPU training experiments and performance benchmarks for both custom and Megatron-LM implementations.
  • Conducted comparative analysis across multiple configurations, including hyperparameter search, optimizer variants, learning rate schedules, and an auxiliary loss–free method, as part of the Swiss AI Initiative.

• Machine Learning Engineering Intern @ Logitech

  Aug 2024 – Feb 2025

  Lausanne, Switzerland

  • Built a pipeline to collect keyboard signal data, followed by preprocessing and feature extraction for downstream ML tasks.
  • Trained lightweight LSTM and FNN models to detect behavioral changes during typing, optimized for limited computational resources.
  • Improved end-to-end user satisfaction through behavior-aware signal analysis and model deployment.

• Student Researcher @ CLAIRe Lab at EPFL

  Feb 2024 – Dec 2024

  Lausanne, Switzerland

  • Developed a custom SFT-based LLM alignment method called Alignment as Translation, and conducted a comparative analysis with state-of-the-art methods such as DPO and ORPO.
  • Evaluated models based on helpfulness, harmlessness, coherence, and truthfulness.
  • Applied LoRA finetuning to assess alignment performance under resource constraints.

• Machine Learning Engineer @ Telescope Labs

  Nov 2022 – Sept 2023

  London, UK

  • Enhanced a production chatbot with probabilistic models, semantic search, and prompt engineering.
  • Deployed custom AI solutions for data-driven analysis in sectors like gaming.

• Student Researcher @ The Scientific and Technological Research Council of Turkey (TUBITAK)

  Jan 2022 – June 2023

  Istanbul, Turkey

  • Built a deep learning system for text transcription using LSTM/Transformer-based decoding with beam search.
  • Developed a page segmentation pipeline for historical documents.
  • Created an interactive GUI for processing and transcribing archival material.

• Data Scientist (Part-time) @ Getir

  March 2022 – Sept 2022

  Istanbul, Turkey

  • Built ML models to improve warehouse efficiency and delivery accuracy.
  • Developed predictive algorithms to identify missing items in shopping baskets.
  • Performed SQL-based data analysis to support business unit decisions and operational improvements.

• Artificial Intelligence Intern @ Huawei Turkey R&D Center

  Aug 2021 – Sept 2021

  Istanbul, Turkey

  • Prepared and processed Named Entity Recognition (NER) datasets.
  • Evaluated pretrained models for integration into the Petal Search engine, Huawei’s in-house search platform.

• Software Engineering Intern @ ASELSAN

  June 2021 – Aug 2021

  Ankara, Turkey

  • Developed a website allowing multiple users to simultaneously watch real-time broadcasts from cameras of their choice.
  • Worked with WebRTC and Kurento for real-time communication.

Projects

• Finding the Right Optimization for Mixture-of-Experts @ EPFL - Machine Learning and Optimization Lab

  Feb 2025 – July 2025

  Comprehensive study on optimization strategies for Mixture-of-Experts models, exploring trade-offs in load balancing, expressivity, and performance.

  • Differentiated learning rate schedules for expert vs. non-expert parameters
  • Tuned auxiliary loss coefficients and explored auxiliary loss–free methods
  • Compared optimizers (AdamW vs. Shampoo) and activation functions (sigmoid vs. softmax)
  • Varied number of experts to assess impact on validation loss, perplexity, and accuracy

  Repo

• Alignment as Translation (ATT) @ EPFL - CLAIRe Lab

  Feb 2024 – Dec 2024

  Novel LLM alignment method treating preference optimization as a translation task—transforming rejected outputs into accepted ones without a reward model.

  • Reframed LLM alignment as neural machine translation
  • Trained on rejected/accepted response pairs using cross-entropy loss
  • Removed need for reward models (simpler & more stable than DPO/ORPO)
  • Implemented Align–Grow iterative improvement cycle
  • Improved helpfulness, harmlessness, and coherence in benchmarks

• Automatic Transcription of Ottoman Documents Using Deep Learning @ Sabancı University – VERİM (Data Analytics Research and Application Center)

  Jan 2022 – June 2023

  Deep learning–based system for automatic transcription of Ottoman Turkish documents written in Arabic-Persian script. Accepted to the 16th IAPR International Workshop on Document Analysis Systems (DAS2024).

  • Developed LSTM/Transformer-based recognition models with beam search decoding
  • Evaluated Word Beam Search with lexicon and n-gram statistics
  • Analyzed lexicon size, coverage, and language modeling impact
  • Achieved competitive character and word error rates on a 6,828-line test set

  PDF

• Zoom for Big Data Analytics: User Interaction Mechanism for Multimodal Interfaces in Collaborative Analytical Sessions @ Sabancı University - BAVLAB (Behavioral Analytics and Visualization Lab

  Sept 2021 – June 2022

  Bachelor's final thesis project: an interactive meeting environment combining Kinect-based hand gesture recognition and multimodal interfaces to enable natural collaboration in online analytical sessions. Applied to the Atlas of Opportunities platform to support decision-making with rich socioeconomic data visualizations.

  • Designed Kinect camera setup for table-projected visual data with gesture control
  • Implemented multiple gesture recognition approaches, resolving environmental constraints
  • Enhanced Atlas of Opportunities for collaborative exploration of socioeconomic and mobility data
  • Supported policymaking and investment planning with interactive spatial analytics

  PDF

• System-Theoretic Analysis of Nano-Communication Signals @ Sabancı University

  Sept 2020 – Feb 2021

  Undergraduate research project as a Student Researcher, exploring molecular communication through the lens of system theory, focusing on enzyme–substrate complex behavior in nanoscale communication channels.

  • Reviewed academic literature on molecular communication and system-theoretic modeling
  • Applied linear regression and neural networks to predict output from enzyme–substrate complex input signals
  • Analyzed and interpreted results to assess prediction accuracy and model applicability

  PDF

Publications

• Automatic Transcription of Ottoman Documents Using Deep Learning

  2024

  Bilgin Tasdemir, E. F., Tandogan, Z., Akansu, S. D., Kızılırmak, F., Sen, M. U., Akcan, A., Kuru, M., Yanikoglu, B.

  Proceedings of the 16th IAPR International Workshop on Document Analysis Systems (DAS 2024) [Accepted]

  PDF