Cornelius Wolff

PHD RESEARCHER · INSIGHT RETRIEVAL FROM STRUCTURED DATA

BSky

Hello! I'm Cornelius Wolff, a PhD researcher at the TRL Lab at the Centrum Wiskunde & Informatica (CWI) and the University of Amsterdam, under the supervision of Madelon Hulsebos and Maarten de Rijke. My work centers on Insight Retrieval from Structured Data, where I investigate AI models and pipelines that can deal with all kinds of structured data like databases and CSV files. I'm furthermore deeply interested in AI in education, In-Context Learning, and efficient machine learning.

Latest publications:

SQALE: A Large Text-to-SQL Corpus Grounded in Real Schemas — AI for Tabular Data Workshop @ Eurips 2025

Are We Asking the Right Questions? On Ambiguity in Natural Language Queries for Tabular Data Analysis — AI for Tabular Data Workshop @ EuRIPS 2025

MinervAI: Using Generative AI to Assist, Not Replace Humans in Peer Review — AI in Science (AIS) 2025 – Copenhagen, Denmark

Education.

2025 – Present

Ph.D. in Computer Science

Centrum Wiskunde & Informatica & University of Amsterdam

Research focus: Insight Retrieval from Structured Data. Supervised by Madelon Hulsebos and Maarten de Rijke at the TRL Lab.

2021 – 2025

M.Sc. in Cognitive Science

Osnabrück University, Germany — GPA: 1.0 (A+)

Focus: Machine Learning, Computational Neuroscience, Ethics of AI. Thesis: Emergence of language in situated environments (Grade: 1.0)

2018 – 2021

B.Sc. in Information Systems

Osnabrück University, Germany — GPA: 2.1

Thesis: Simulation environment for analyzing the spread of SARS-CoV-2 using ML agents (Grade: 1.0)

Work Experience

PhD Student, CWI & University of Amsterdam (Mar 2025 – Present)

Junior Researcher, German Research Center for AI (DFKI), Osnabrück (Mar 2022 – Mar 2025)

Research Assistant, Elia Bruni's Lab, Osnabrück University (Apr 2023 – Mar 2025)

Research Assistant, Distributed Systems Lab, Osnabrück University (Jun 2021 – Mar 2023)

Technical Skills

Python PyTorch TensorFlow SQL TypeScript LLMs RAG Reinforcement Learning Vector Databases Git LaTeX

Research.

Insight Retrieval from Structured Data

Investigating AI models and pipelines that extract meaningful insights from structured data sources such as databases, CSV files, and tabular datasets.

Small & Interpretable Language Models

Developing compact, interpretable language models that achieve high performance while remaining transparent and computationally efficient.

In-Context Learning & Reinforcement Learning

Applying ICL principles to image classification, tabular data, and other structured domains; studying emergent communication in multi-agent RL systems and situated environments.

Publications.

SQALE: A Large Text-to-SQL Corpus Grounded in Real Schemas

AI for Tabular Data Workshop @ Eurips 2025 — 2025

Introduction of SQALE, a large-scale semi-synthetic text-to-SQL dataset grounded in real relational schemas, supporting research on scalable and generalizable NL2SQL models.

READ THE PAPER ↗

Are We Asking the Right Questions? On Ambiguity in Natural Language Queries for Tabular Data Analysis

AI for Tabular Data Workshop @ EuRIPS 2025 — 2025

A conceptual framework for characterising ambiguity in natural-language queries over tabular data, arguing for cooperative query specification and analysing 15 common tabular-data benchmarks.

READ THE PAPER ↗

MinervAI: Using Generative AI to Assist, Not Replace Humans in Peer Review

AI in Science (AIS) 2025 – Copenhagen, Denmark — 2025

A position paper arguing for constrained, verifiable uses of LLMs in peer review and introducing MinervAI—an open-source tool that supports citation verification and argumentation mapping while preserving human judgment.

READ THE PAPER ↗

How well do LLMs reason over tabular data, really?

4th Table Representation Workshop - ACL 2025 — 2025

Analysis of the reasoning capabilities of LLMs over tabular data, presented at the 4th Table Representation Workshop at ACL 2025.

READ THE PAPER ↗

PictSure: Pretraining Embeddings Matters for In-Context Learning Image Classifiers

arXiv preprint arXiv:2506.14842 — 2025

PictSure: demonstrating the importance of pretraining embeddings for in-context learning image classifiers.

READ THE PAPER ↗

Bidirectional Emergent Language in Situated Environments

arXiv preprint arXiv:2408.14649 — 2024

Research on bidirectional emergent language in situated environments, exploring how agents develop communication systems in interactive settings.

READ THE PAPER ↗

GRASP: A Novel Benchmark for Evaluating Language Grounding and Situated Physics Understanding in Multimodal Language Models

Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence, IJCAI-24 — 2024

GRASP benchmark for evaluating language grounding and situated physics understanding in multimodal language models, presented at IJCAI-24.

READ THE PAPER ↗

A Review of Nine Physics Engines for Reinforcement Learning Research

arXiv preprint arXiv:2407.08590 — 2024

A comprehensive review of nine physics engines for reinforcement learning research, analyzing their capabilities and suitability for RL applications.

READ THE PAPER ↗

A New Approach on Estimating Germany's Mobile Broadband Coverage based on Crowdsourced Data

Mobile Communication-Technologies and Applications; 27th ITG-Symposium — 2023

A novel approach for estimating mobile broadband coverage in Germany using crowdsourced data, presented at the 27th ITG-Symposium on Mobile Communication.

READ THE PAPER ↗

Teaching & Talks.

Study Project: LLMs for Science

Master's Study Project — Osnabrück University · Osnabrück, Germany · 2024

Managed and organized a study project focused on “Supporting the Review Process with AI” - an AI-driven system developed by master’s students to assist with the academic reviewing process.

VIEW DETAILS ↗

Emergent Behavior in Multi-Agent Systems (EBIMAS)

Master's Study Project — Osnabrück University · Osnabrück, Germany · 2022

Started and organized a study project to provide a platform and course setting for cutting edge Reinforcement Learning and emergent behavior research.

VIEW DETAILS ↗

SQALE: A large Text-to-SQL dataset with Realistic Database Schemas

Seminar Talk — TRL Seminar, CWI · Amsterdam, Netherlands · January 23, 2026

YouTube RecordingIn this talk at the TRL Seminar Series at CWI, I presented SQALE, a large-scale dataset aimed at advancing text-to-SQL systems through more realistic training and evaluation data.Natural language interfaces for databases rely on text-to-SQL models to translate user questions into executable SQL quer...

WATCH RECORDING ↗

How well do LLMs reason over tabular data?

Online Talk — AlphaXiv · Amsterdam, Netherlands · September 3, 2025

Youtube RecordingIn this presentation at AlphaXiv, I talked about my paper on “How well do LLMs reason over tabular data, really?”, which is about whether general-purpose Large Language Models can effectively reason over tabular data. We identified flaws in current evaluation methods and proposes an LLM-as-a-judge ...

WATCH RECORDING ↗

Can We Create Green AI? - Podcast Interview

Podcast Interview — Kaleidoscience: Conversations on Cognitive Science · Osnabrück, Germany · December 19, 2024

Podcast EpisodeIn the 12th episode of the podcast “Kaleidoscience: Conversations on Cognitive Science,” I discussed with Sönke Lülf and Elisa Palme about the sustainable use of Artificial Intelligence and how we can create green AI.Topics: The energy footprint of AI models Sustainable data management in companies ...

WATCH RECORDING ↗

Cornelius Wolff

Latest publications:

Education.

Ph.D. in Computer Science

M.Sc. in Cognitive Science

B.Sc. in Information Systems

Work Experience

Technical Skills

Research.

Insight Retrieval from Structured Data

Small & Interpretable Language Models

In-Context Learning & Reinforcement Learning

Publications.

Teaching & Talks.

Contact.