Tassilo J. Klein, Ph.D.

Director of ML Research · Principal Scientist · SAP AI

I work on foundation models, agentic AI, and generative interfaces — with a focus on making large-scale AI systems actually work on the messy, structured data that enterprises run on.

About

I lead ML research at the SAP AI CTO Office, where my team works at the frontier of foundation models, agentic AI systems, and Generative UI. My research spans the full stack: from representation learning and self-supervised pre-training to autonomous agent runtimes and AI-native interface generation.

I’m particularly interested in the coordination economics of agentic AI — how protocol-mediated agent systems change the cost structure of enterprise software, and what that implies for how we build and deploy AI at scale. I’ve open-sourced MCPRuntime, a high-performance local runtime for autonomous agents, available on PyPI.

Previously: postdoc at MIT CSAIL and Harvard Medical School (representation learning, multimodal medical imaging). Ph.D. summa cum laude from TU Munich. Member of ELLIS.

Research

Area	Focus
Agentic AI	Autonomous agent runtimes, MCP-based tool calling, multi-agent orchestration, self-growing tool libraries
Foundation Models for Structured Data	Tabular pre-training, relational reasoning, enterprise-scale data representation
Generative UI (GenUI)	AI-driven dynamic interface generation, adaptive UX, intent-to-UI synthesis
Knowledge-Augmented LLMs	RAG, privacy-preserving fine-tuning, domain adaptation
Representation Learning	Contrastive and self-supervised methods, multimodal alignment

📄 Selected Publications & Projects

[2026.03] - Released MCPRuntime — An open-source, high-performance local runtime for autonomous agents implementing programmatic tool calling and self-growing tool libraries.

[2026.02] - New pre-print available on how agentic AI reshapes enterprise boundaries

The Headless Firm: How AI Reshapes Enterprise Boundaries

[2025.11] - A paper accepted (spotlight) at the EurIPS’25 Workshop on AI for Tabular Data

SALT-KG: A Benchmark for Semantics-Aware Learning on Enterprise Tables

[2025.07] - A paper accepted at the SIGMOD’25 Workshop on Data Management for End-to-End Machine Learning

Table Dissolution: Adding Salt To Your Data

[2025.07] - SALT officially integrated into RelBench.

[2025.05] - New pre-print available on foundation models for tabular data in enterprises

[2025.05] - Paper accepted at ACL 2025 — Contrastive Perplexity for Controlled Generation: An Application in Detoxifying Large Language Model

[2024.10] - Two papers accepted at the NeurIPS’24 Table Representation Learning Workshop

SALT: Sales Autocompletion Linked Business Tables Dataset - pre-print
PORTAL: Scalable Tabular Foundation Models via Content-Specific Tokenization - pre-print

[2023.05] - Paper accepted at ACL 2023 on low-shot contrastive learning of sentence representations.

[2022.02] Paper accepted at ACL 2022 on self-supervised sentence representation learning

[2021.08] Paper at EMNLP 2021 on Contrastive Language Model Refinement for Commonsense Reasoning

[2021.08] Paper at EMNLP 2021 on Contrastive Self-Supervised Learning for Commonsense Reasoning

[2021.04] Acceptance of co-organized at ICML 2021 workshop on Self-Supervised Learning for Reasoning and Perception

[2021.02] Paper accepted at IPMI 2021 on self-supervised representation learning for medical imaging (acceptance rate 30.0%)

[2020.09] Presentation on commonsense reasoning in AI

[2020.04] Paper accepted at ACL 2020 on contrastive self-supervised commonsense reasoning (acceptance rate of 17.6%)

[2020.02] Paper accepted at NeuroImage

[2019.10.20] Paper on Multi-Domain Learning accepted at ICCV 2019 (acceptance rate 25.0%)

[2019.05.14] Short-paper on commonsense reasoning accepted at ACL 2019 (acceptance rate 18.2%)

[2019.02.25] Paper accepted at CVPR 2019 (acceptance rate 25.2%)

[2017.02.01] Paper accepted at NeuroImage

Open Source

I believe frontier research should ship. If you’re working on agentic AI runtimes, tabular foundation models, or generative interfaces — reach out or open an issue.

Service & Mentorship

Program committee / reviewer: ACL, EMNLP, CVPR, NeurIPS
Mentored 15+ researchers and Ph.D. students across academia and industry
Co-organizer: ICML 2021 Workshop on Self-Supervised Learning for Reasoning and Perception

Alumni

Former interns who have gone on to research roles at top labs and faculty positions:

Enrico Fini — Member of Technical Staff at Microsoft AI (previously Research Scientist at Apple Machine Learning Research)
Stefan Lionar — Ph.D. Student in Computer Science at National University of Singapore (NUS) & Sea AI Lab
Jan Nikolas Morshuis — Doctoral Researcher at the University of Tübingen (Cluster of Excellence “Machine Learning” / Tübingen AI Center)
Aiham Taleb — Senior Applied Scientist – Generative AI at Amazon Web Services (AWS) / Generative AI Innovation Center
Artur Speiser — AI Research Scientist at Exscientia, formerly Ph.D. in Machine Learning in Science at University of Tübingen
Jannik Wolff — Machine Learning Research Scientist & Research Associate at BIFOLD, Ph.D. Candidate at TU Berlin (Machine Learning Group)
Daniel Dorda — Scientific Software Engineer at uniqFEED and researcher/teaching assistant at ETH Zürich (Computer Graphics / Computer Science)
Max Bain — Research Scientist at Google DeepMind, previously Member of Technical Staff at Reka and Ph.D. at VGG, University of Oxford
Mahdyar Ravanbakhsh — (Principal) Research Scientist at Zalando SE in Berlin
Oleksiy Ostapenko — Research Scientist at ServiceNow AI Research (Foundation Models Lab); Ph.D. from MILA / Université de Montréal
Colin Samplawski — Ph.D. Student at UMass Amherst (REML group) and Advanced Computer Scientist at SRI International (NuSCI Research Group)
Mihai M. Puscas — Computer Vision Researcher at SPORTTOTAL.TV (previously Huawei Research, Dublin)
Sandro Pezzelle — Assistant Professor in Responsible AI at ILLC, University of Amsterdam; Scientific Advisor at IVADO Labs
Frederik Pahde — Research Associate / Ph.D. Candidate at Fraunhofer HHI & TU Berlin (previously Applied ML Scientist at Amazon)
Shailza Jolly — Research Scientist at Amazon Alexa AI in Berlin (Ph.D. in Generative AI from TU Kaiserslautern / DFKI)
Denis Dushi — Machine Learning Engineer at Amazon
Vadim Tschernezki — Ph.D. Student in Computer Vision at the Department of Engineering Science, University of Oxford (VGG)
Robin C. Geyer — Ph.D. Student at the Institute for Machine Learning, ETH Zürich

_{Last updated — March 2026}