16th Munich NLP Meetup featuring trustful GenAI applications

Dear NLP Munich Community members,

We would like to invite you to our 16th NLP meetup 🥳 This time it'll be all about LLMs again (what else?😉🙈) with two exciting talks and an open discussion with you all.

Let us start again gathering the excellent NLP community in Munich and having great topics, interesting discussions and fine networking 🚀

Both talks will address the application of LLMs in sensitive use cases and domains with specific needs to 1) mitigate risks and implement trustful solution 🔥, and 2) to gain robustness in production 🔥.

After the talks, we would like to have an open discussion with you all to reflect about your challenges and potential solutions in industrialization of your Generative AI applications 🤗🤗.

=== Talks ===

Here are the talks and their brilliant speakers:

Generative Editorial AI at Ippen Digital — How to take journalistic accountability for Large Language Models

• Speakers:
• Alessandro Alviani (Product Lead AI Team, Ippen Digital)
• David Kaumanns (Technical Lead AI & Machine Learning, Ippen Digital)
• Abstract: Editorial journalists require creation tools that are reliable and accountable. How to proprietary LLMs satisfy these needs? Can decades of article texts be leveraged constructively? Ippen Digital's AI Team provides exclusive insights into their progress and lessons learned.

The Case of Virtual Havruta: Practical Engineering Challenges and Directions in Industrializing Retrieval-Augmented Generation for Large Language Models

• Speaker: Paul Yu-Chun Chang (appliedAI Initiative GmbH)
• Abstract: Retrieval Augmented Generation (RAG) has emerged as a crucial strategy over the past year for enhancing the trustworthiness of Large Language Models (LLMs). A RAG system entails identifying documents that semantically align with a query and subsequently leveraging a Large Language Model, such as ChatGPT, to extract the correct answer from these documents. Despite its potential, RAG systems confront challenges inherent to information retrieval systems and their dependence on LLMs. In this presentation, we delve into the practical engineering challenges faced by companies embarking on their RAG journey, utilizing an intriguing and sophisticated case in the open-source domain—Virtual Havruta. This recent use case, stemming from a collaborative effort involving TUM Venture Labs, Sefaria, and the appliedAI Initiative, employs RAG techniques to develop a study companion tailored for users with a keen interest in delving deeply into the highly specific domain of Jewish scripture study. Our discussion will encompass common issues encountered, ranging from the chunking strategy in the pre-retrieval stage to query optimization during retrieval and re-ranking in the post-retrieval stage. With the objective of fostering interactive discussions with the audience, we seek to inspire potential directions for the industrialization of RAG systems.

=== Agenda ===

• 18:00: Doors open
• 18:45 – 19:00: Welcome and Introduction of our host Ippen Digital
• 19:00 – 19:45: Talk 1: Alessandro Alviani, David Kaumanns (both Ippen Digital)
• 19:45 – 20:00: [Break] Networking and refreshment
• 20:00 – 20:45: Talk 2: Paul Yu-Chun Chang (appliedAI Initiative GmbH)
• 20:45 – 21:00: Open Discussion: Your Challenges and your Solutions of industrialing GenAI applications
• 21:00 - 22:00: [Networking] Get-together with sponsored food and drinks

=== Sponsor ===

And eventually, we want to thank Ippen Digital for hosting this evening. If you don’t know Ippen Digital yet, you will learn more in a short introduction at the beginning.

Looking forward to meeting you!

Best Bernhard, Christoph, Kristian and Kun