Jonas Klesen

- Built end-to-end LLM systems to extract structured knowledge from scientific literature. - Fine-tuned open-weight LLMs (1B-405B) using SFT and reinforcement learning. - Scaled self-hosted LLM inference to billions of tokens. - Built cost-efficient pipelines processing millions of scientific PDFs. - Developed rigorous LLM output evaluation frameworks with key metrics. - Created full-stack annotation and quality-assurance platforms for users. - Presented research via tech talks and executive presentations.

Master thesis: "Guided Pruning for Resource-Efficient Specialized Transformer Models", supervised by Prof. Isabel Valera. Grade: 1.0

B.Sc. thesis: "Black-Box Adversarial Attacks in Machine Learning", supervised by Prof. Bernt Schiele at Max Planck Institute for Informatics. Grade: 1.0