Peter Izsak builds language systems that make research useful in practice.
Practical language-model research across evaluation, retrieval, efficient training, and healthcare applications, shaped for systems that work outside the lab.
Selected work
Recent papers and one durable signal.
- International Conference on Learning Representations (ICLR 2025) HELMET: How to Evaluate Long-Context Language Models Effectively and Thoroughly
- Preprint RAG Foundry: A Framework for Enhancing LLMs for Retrieval Augmented Generation
- The 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP 2021) How to Train BERT with an Academic Budget
- 2019 Fifth Workshop on Energy Efficient Machine Learning and Cognitive Computing - NeurIPS Edition (EMC2-NIPS) (2019): 36-39. Q8BERT: Quantized 8Bit BERT
Current focus
Evaluation, retrieval, and efficient model adaptation.
The throughline is practical language-model research: making long-context evaluation sharper, retrieval systems easier to improve, and model training more efficient under real constraints.