Ruşen Birben
AI and Data Engineering graduate from Istanbul Technical University and independent researcher with METU’s Applied NLP Group. Research interests center on LLM reliability and interpretability, with experience in Turkish NLP, machine unlearning, and LLM applications for finance.
EDUCATION
B.Sc. AI and Data Engineering
Istanbul Technical UniversityGraduated at 09.07.2025
Electronics and Communication Engineering
Istanbul Technical UniversityTransferred to AI & Data Engineering
SCHOLARSHIPS & AWARDS
1416 YLSY Scholarship · M.Sc. & Ph.D.
Türkiye Ministry of National EducationAwarded government funding for overseas master’s study followed by doctoral study, covering tuition fees, a monthly living stipend, health insurance, travel, language-exam fees, and study supplies. Following completion of the funded studies, will return to Türkiye as an assistant professor.
Ranked 677th of 6,807 teams (top 10%) in Kaggle’s simulation competition, developing and evaluating partially observable game-playing agents with replay-based behavioral cloning, belief-aware Monte Carlo planning, and self-play.
RESEARCH EXPERIENCE
Contribute to data creation and evaluation methodology for the SIGTURK 2027 benchmark on Turkic languages and cultural understanding with Çağrı Toraman. Support LlamaTurk2 through training-data collection and pretraining experiments. Lead METU NLP Connect, producing expert interviews.
Investigated software-log analysis for predictive maintenance; developed prototypes using HDFS logs.
Researched financial LLM applications involving sentiment analysis, tool use, reasoning, forecasting, and retrieval-augmented generation. Contributed experiments to the Kaira undergraduate capstone.
PUBLICATIONS & RESEARCH CONTRIBUTIONS
Exact First, Then Fast: An Event-for-Event C++ Rebuild of the Agenthon ABIDES Market
Agenthon Workshop @ NeurIPS 2026 - Accepted for in-person poster presentationGöktürk Batın Dervişoğlu, Muhammed Faruk Aytin, Muhammed Rüşen Birben.
Rebuilt the Agenthon ABIDES market simulator in C++, preserving event ordering, random-number sequences, and reference outputs. Validated exact agreement across 6,900 randomized comparisons and achieved a 1,299× simulation-loop speedup on the largest public benchmark, measured on the same laptop.
Atlanta, USA, December 12, 2026.
Turkish Cultural Knowledge Dataset
MRL Workshop @ EMNLP 2026 - Accepted dataset contribution; joint benchmark paper in preparationLed the source-grounded agent pipeline and data preparation for a Turkish cultural knowledge dataset spanning 15 domains.
Screened 1,200 candidate question-answer pairs; selected 500 for sociological review by Büşra Tufan and prepared the final 300-pair submission.
Contributed to English translation review under the research guidance of Çağrı Toraman.
SELECTED PROJECTS
Kaira
Undergraduate Capstone: Financial Reasoning LLMFine-tuned Qwen-based LLMs with GRPO, Unsloth, and LoRA; ran FinQA and GSM8K experiments on reward design and financial reasoning. Developed an Ollama-based chat application.
Tobor
Autonomous Warehouse RobotImplemented the box pickup/drop-off mechanism in a team-built ROS2 warehouse robot with navigation and YOLO perception; shared data annotation and supported other algorithms.
AIZheimer
Machine Unlearning in Stable DiffusionAdapted Forget-Me-Not-based unlearning for Stable Diffusion 2.1. Implemented attention controllers and explored loss objectives; qualitatively assessed concept removal and retention of unrelated generation capabilities.
Biting The Bytes
Turkish Diacritic RestorationLed ByT5-small training and hyperparameter tuning with PEFT/LoRA for byte-level Turkish diacritic restoration. Shared debugging and Hugging Face Spaces deployment with a teammate.
PROFESSIONAL EXPERIENCE
Co-founded Fiction Studios at Bilkent Cyberpark; led technical development and coordinated three additional contributors. Built a multilingual LLM-powered museum and tourism guide, alongside customer and investor discussions.
Developed news-sentiment pipelines, internal financial LLM tools, web integrations, and a mobile voice assistant at a Phillip Capital subsidiary. Worked on retrieving current context from fragmented sources and PDF reports.
TECHNICAL SKILLS
- Languages
- Python · JavaScript/TypeScript · SQL · C/C++
- Frameworks & Tools
- PyTorch · TensorFlow · Transformers · PEFT / LoRA · Unsloth · Scikit-learn · Pandas · NumPy · ROS2 · YOLO
- Specialties
- NLP · LLMs · RAG · Machine Unlearning · Model Evaluation
COURSES
Deep Learning Specialization ↗
DeepLearning.AI · CourseraNeural networks, CNNs, sequence models, optimization, and structuring machine learning projects.
AWS Cloud Technical Essentials ↗
AWS · CourseraCloud architecture and AWS services including EC2, Lambda, S3, RDS, and DynamoDB.
LANGUAGES
Turkish
Native
English
TOEFL iBT 109/120 (November 8, 2025)
INTERESTS
rusenbirben.comCoffee
V60 - Aeropress - Moka Pot
Books
Philosophy - AI/CS - Finance
Photography
Street Photography - Double Exposure - Bird Photography
Anime & Manga
Watcher - Reader - Collector