Save the Date!

​We’re thrilled to announce our next meetup, hosted in collaboration with SAP at their office, Friedrich-Ludwig-Bauer-Str. 5, 85748 Garching bei München on Tuesday, August 25, 2026!

⚠️ Please register on Luma: https://luma.com/b6n5x3by — a Meetup RSVP alone does not reserve a place. Attendance is confirmed only after registration and approval through Luma, since capacity at the SAP office is limited.

RSVP

About the Event

​We’re thrilled to announce our next meetup, hosted in collaboration with SAP! Join us for an evening bringing together research and industry perspectives on long-context LLMs, information retrieval, knowledge graphs, and autonomous AI agents.

What role do knowledge graphs play as LLMs get longer context windows and AI agents move into production? Expect two technical talks, food and drinks, and plenty of time to meet researchers, engineers, students, and other members of Munich’s NLP and AI community.

Agenda - TIMETABLE:

• ​17:30 | Doors open
• 17:50 – 18:00 | Welcome and introduction (SAP, MunichNLP)
• 18:00 – 18:45 | Talk 1: “Custom AI in Action”
• ​18:45 – 18:50 | Short break ​ - 18:50 – 19:35 | Talk 2: “Needle in a Knowledge Graph”
• ​19:35 – 21:00 | Networking & drinks 🍕

Talks

• Riswan Lathif: Custom AI in Action

  SAP’s customers are varied and cover a lot of industries. Thus they also have a need to build their own custom solutions to make the best out of SAP’s standard software. This presentation shows how custom AI solutions can help leverage the available AI tools to add business value and discusses how we move beyond promising AI prototypes and turn autonomous agents into productive, reliable workflows.

  Riswan is a principal software architect at Customer Industry Solutions, who specializes in building custom AI solutions. He worked in startups and the German space center DLR before joining SAP, and is an advocate of using AI responsibly to its full potential.

• Jakob Sturm: Needle in a Knowledge Graph

  Do we still need retrieval when LLMs can digest increasingly large corpora through ever-growing context windows? This project investigates whether LLMs can reliably identify relevant information within large inputs by introducing a challenging benchmark based on knowledge graphs. The research examines how model performance changes as input size increases and how the ordering of information within the context affects an LLM’s ability to locate the relevant information.

  Jakob Sturm is a PhD researcher at TUM’s Social Computing Group and BMW’s Innovation & Emerging Technologies department, supervised by Georg Groh. His research focuses on Information Retrieval and Retrieval-Augmented Generation, particularly on understanding when RAG is the right approach and when other methods may be more suitable. At BMW, his industry focus is on domain-adapted RAG for quality-management processes.