Projects
Research, publications, and software
Research
Currently optimizing retrieval models for AI-agent consumption, addressing how agents ingest data differently and are sensitive to noise while existing search interfaces and interactions remain un-optimized for agentic use.
Publications
- Yilmazel O., Zamani H., Allan J. Exploring and Improving Cross- and Multi-Domain Personalized Question Answering via a Dual Retrieval Augmentation Approach. The 49th International ACM SIGIR Conference on Research and Development in Information Retrieval, Short Papers Track; 2026. SIGIR 2026 Paper
- Yilmazel O., et al. Does Retrieval Augmented Generation Collapse When It Retrieves AI-Generated Documents? GroundLM; 2026; preprint. openreview.net/forum?id=JHwIkYmtxb
- Yilmazel O., Zhu A., Navarrete P., Pogorelov S., Partridge V. iNatator: Obtaining Expert Feedback on Species Ranges. New England Computer Vision Workshop (NECV); November 2024; Yale University, New Haven. Poster
- Yilmazel O., McGrady R., Partridge V., Zuckerman E. YouTube Topic Modeling. Computational Social Science Poster Session 2025; May 2025; UMass Amherst. 🏆 Winner of the People's Choice Award Poster
- Myers J. R., et al. Developing an equitable machine learning-based music intervention for older adults at-risk for Alzheimer's: Pilot findings for algorithm development and validation. JMIR Aging; October 2025. Contributor Poster
Projects
OpenResearcher — Funding and collaboration platform
- Building a platform that enables academia and industry to find research collaborators, equipped with agentic search functionalities that help industry companies discover labs and academics to fund.
- Developed the full stack with FastAPI and Next.js, including a custom retrieval server offering lexical and semantic retrievers that operate independently or in a hybrid configuration.
- Integrated retrieval with agentic search and automated report generation using LangChain.
Art-GAN — Generating natural patterns with generative adversarial networks
- Photographed real-world natural patterns, plants, flowers, and tree bark, and trained a GAN to synthesize new organic patterns, exploring how models perceive the visual structures found in nature.
- Probed the model's intermediate layers and inspected layer activations to surface novel and visually compelling artifacts beyond the final outputs.
- Early Examples
Natural Language Playlist Querying
Fall 2025
- Designed and implemented a natural language playlist retrieval system, developing custom song embeddings and a dual-encoder retriever.
- Processed 100,000 playlists from Spotify, leveraging PyTorch and HuggingFace Transformers to train retrieval models.
Scoundrel — Solo CLI Dungeon
I was bored one night
- A simple python CLI game based on the scoundrel card game.
- Why? CLI is fun and now I don't have to carry cards or shuffle.
OverMath — A Math Cooking Game
Fall 2024
- Co-developed a Unity-based educational game inspired by Overcooked; implemented Adversary AI, player controls, and projectile physics.
- Contributed core gameplay mechanics as part of a 5-person team.