I study how people seek and discover information—and make systems respond better. My work spans search, recommendation, and rigorous evaluation.
Connect via Email, LinkedIn, or X; explore my work on Google Scholar and GitHub; for collaborations, my CV.
About Me
I’m an Applied Scientist intern at Amazon Search and a PhD in Computer Science from TU Wien. I study how people seek and discover information—and build systems that respond better. My work sits where search and recommendations meet, with a focus on making results feel relevant, consistent, and fair for real users.
I care a lot about rigor and reliability: clear problem framing, trustworthy evaluation, and reproducible workflows. I like turning research ideas into clean, well-documented systems that teams can maintain and grow with confidence. Along the way, I’ve collaborated with industry and academia and published 30+ papers (15 first-author, 600+ citations). If you’re into search, recommender systems, and rigorous evaluation, we’ll have plenty to talk about..
Experience
May 2025 - Oct. 2025, Applied Scientist Intern
Amazon Search, Luxembourg
As an Applied Scientist Intern at Amazon Search, I advanced query understanding by developing a unified multi-task model based on a Mixture-of-LoRA Experts architecture. My work emphasized context-aware expert routing to improve consistency across signals, balance task interactions, and enable efficient adaptation under compute constraints. In addition, I established a reproducible experimentation pipeline, introduced new evaluation metrics, and delivered actionable insights to science and product teams.
▸Details
- Unified query understanding models into one multi-task model using a Mixture of (LoRA) Experts, applying smart routing to foster positive interference and mitigate negative interference.
- Built a reproducible experimentation pipeline with PyTorch Lightning, AWS SageMaker, and MLflow, including hyperparameter optimization, ablations, and clean experiment tracking.
- Defined and implemented consistency metrics alongside accuracy, and automated evaluation/reporting to guide model and product decisions.
- Maximized efficiency under compute constraints using LoRA-based experts, vectorization, caching, and right-sized SageMaker runtime parameters.
- Authored design docs and presented to science/product stakeholders; prepared a reproducibility package and conference-style write-up; refactored code for maintainability and handover
Oct. 2018 - Jan. 2024, Research Assistant
TU Wien, Austria
As a Research Assistant at TU Wien, I conducted extensive research in User Modeling, Recommender Systems, and Information Retrieval, collaborating with researchers from diverse backgrounds and open-sourcing impactful tools and artifacts. I worked closely with industry partners, aligning research with practical needs, and shared my expertise in the classroom—lecturing, assisting in courses, and mentoring students to support their academic and professional growth.
▸Details
- Published 30+ research papers (15 as first author), with 600+ citations, by conducting extensive research in Information Retrieval (IR) and Recommender Systems (RecSys) and collaborating with diverse research teams.
- Increased accessibility and impact of research tools by developing and open-sourcing IR and RecSys demos, repositories, and tools, resulting in 20+ GitHub stars and 10+ forks.
- Led industry collaboration and grant writing efforts, securing funding (thousands of euros) and maintaining partnerships with companies like Geizhals, Falter, YouKnowMeBest, OIDA, and Tripmakery. Gathered industry requirements, translated them into applied research plans, and aligned lab goals with partner business objectives through regular meetings and updates.
- Independently led and transitioned a 70-student Social Network Analysis course online during COVID-19, covering all lectures, Zoom sessions, and assessments in the primary lecturer’s absence due to maternity leave.
- Enhanced inclusivity in a 130-student Recommender Systems course by shifting practicals to auto-graded Jupyter Notebooks and managing Jupyter Lab resources, ensuring access for students with limited technology.
- Mentored 30 students annually in Innovation course, guiding student teams of 5 through ideation, business plan development, mockups, pitches, and financial projections, including cost, revenue, and break-even calculations.
Sept. 2014 - Sept. 2017, Data Engineer
Wingpaper, Austria
At Wingpaper, a startup aiming to create a "YouTube for documents," I focused on enhancing platform scalability and accessibility. My role involved ingesting data—primarily PDFs—transforming it into searchable and presentable formats. To make content more engaging, I developed a hybrid recommender system that used both content-based and collaborative filtering, boosting recommendation quality and improving user engagement metrics.
▸Details
- Boosted system scalability by designing and implementing a microservice architecture with Redis and RabbitMQ.
- Increased document processing speed by 12x (from 2 to 24 pages per second) by developing and scaling microservices for indexing, information extraction (using Elastic and Python), and PDF-to-HTML5 conversion (pdf2htmlex).
- Improved recommendations, achieving a .25 improvement in NDCG@5 and a 5% increase in CTR by building a hybrid recommender system that combines content-based (keyword extraction) and collaborative filtering (interaction data) techniques using Python.
Apr. 2008 - Aug. 2014, Web Developer
hyperCMS , Austria
As a Web Developer at hyperCMS, I developed scalable, customized web solutions for global clients and built a comprehensive e-learning platform. I also provided hands-on training to ensure effective platform use across diverse user groups.
▸Details
- Developed custom modules and templates using HTML, CSS, JavaScript, PHP, and MySQL for clients in 20+ countries, supporting over 500 employees. The resulting webpages served millions of customers.
- Built a fully integrated e-learning platform within the CMS, featuring custom templates for quizzes, automated grading, and certification, used by UNIQA Group across 17 countries.
- Led training sessions for 200+ diverse CMS users (tech and non-tech, young and senior), delivering one-on-one and seminar-style sessions, ensuring effective use of the platform as new features were continuously developed.
Research
IR & RecSys Evaluation
My research in Information Retrieval and Natural Language Processing has focused on developing robust evaluation methodologies across multiple test collections, with applications in both IR and Recommender Systems. This work is inspired by the need for more reliable, multi-task evaluation techniques that demonstrate the accuracy and applicability of these technologies.
▸Publications
- 2023 - ACL
Ranger: A Toolkit for Effect-Size Based Multi-Task Evaluation - repo
Mete Sertkan, Sophia Althammer, Sebastian Hofstätter
- 2023 - RecSys, PERSPECTIVES
Exploring Effect-Size-Based Meta-Analysis for Multi-Dataset Evaluation - repo
Mete Sertkan, Sophia Althammer, Sebastian Hofstätter, Peter Knees, Julia Neidhardt
- 2022 - CIKM
Introducing Neural Bag of Whole-Words with ColBERTer: Contextualized Late Interactions using Enhanced Reduction - repo
Sebastian Hofstätter, Omar Khattab, Sophia Althammer, Mete Sertkan, Allan Hanbury
- 2021 - TREC
TU Wien at TREC DL and Podcast 2021: Simple Compression for Dense Retrieval
Sebastian Hofstätter, Mete Sertkan, Allan Hanbury
News Recommendations - Sentiments, Emotions, and Diversity
In this research trajectory, I focus on enhancing news recommendation systems by incorporating sentiment and emotional cues, while exploring their effects on diversity in recommendations. This work highlights both the benefits and risks of sentiment-based models in promoting diverse content.
▸Publications
Tourism Recommendations - Preference Elicitation and User Profiling
In this research, I explored tourism destination characteristics by leveraging gamified image-based strategies to uncover implicit user preferences. The focus was on understanding and predicting user needs through subtle preference elicitation techniques.
▸Publications
For an extensive list of publications visit Google Scholar.
Education
May 2025, PhD in Computer Science
TU Wien, Austria
▸Details
- Thesis: Leveraging the Subtle: Hidden Factors in Recommender Systems.
- Relevant Coursework: Recommender Systems, Deep Learning, Research Methods.
- GPA: 3.89
June 2018, Master’s in Computer Science and Economics
TU Wien, Austria
▸Details
- Thesis: Classifying and Mapping e-Tourism Data Sets.
- Relevant Coursework: Machine Learning, Software Engineering & Project Management, Econometrics.
- GPA: 3.71
July 2011, Bachelor’s in Computer Science and Economics
TU Wien, Austria
▸Details
- Thesis: Gesture-Based Interfaces.
- Relevant Coursework: Programming, Algorithms & Data Structures, Mathematics, Statistics & Probability.
- GPA: 2.81
Certificates
Issued by DeepLearning.AI
Issued by IBM
Issued by Google
Issued by DeepLearning.AI and Stanford University
Issued by Imperial College London
Issued by DeepLearning.AI
Language Skills
Spoken Language
German (native), Turkish (native), English (full professional).
Coding Language
Python, JavaScript, R, Java.
Awards
Recognized as one of 16 nominees for best reviewer out of 264 reviewers, demonstrating a commitment to research quality.
Pitched ImageTwin, a tool to combat plagiarism and data fabrication, and won a €5,000 scholarship with workplace incubation at TUWi2 (TU Wien's startup incubator).
Awarded Best Article for “ What is the Personality of a Tourism Destination?” published in Information Technology & Tourism, selected as top article out of 31 submissions.
Won Best Master’s Thesis Award for “ Classifying and Mapping e-Tourism Data Sets”, selected out of 27 thesis submissions.
Academic & Voluntary Services
Jan. 2020 – Oct. 2023, Program Committee Member
Reviewed submissions for major conferences and journals, including ENTER, CIKM, JITT, RecSys, SIGIR, and UMAP, contributing to the quality and rigor of academic discourse in information retrieval and recommender systems.
July 2022, Streaming & Broadcasting Chair, UMAP 2022
Led online session management and video production for the UMAP 2022 Conference, coordinating with a team of four to ensure smooth online participation for all attendees.
Enhanced accessibility of Soviet Holocaust film documentation by implementing an OCR-Translate-Index pipeline and conducting a comparative study, selecting DeepL as the optimal tool for translating caption sheets and reports.
Launched DIGHUM during COVID by setting up a website, managing bi-weekly virtual lectures, and creating a YouTube channel (100+ videos, 1300+ subscribers). Supported the DIGHUM manifesto with 1000+ signatures and coordinated four on-site workshops.
