What a Day! Tag der Deutschen Einheit 2025 💪🤖

At Tag der Deutschen Einheit (Germany's Unity Day) celebrations in Saarbrücken, our department showcased Mentalytics, an AI system based on Large Language Models (LLMs) that predicts participants' mental states and detects actual pain and effort to perform exercises. Over 220 participants tested the privacy-first technology, which runs entirely on local devices, in an interactive demo that proved both popular and successful.

Under the motto „Zukunft durch Wandel“ around 400,000 people visited the festival miles in Saarbrücken from 2-4 October for the Day of German Unity celebrations – and Smart Service Engineering department, DFKI led by Prof. Wolfgang Maaß was right in the heart of it all!

When AI Knows How You Feel💪🤖
At the Tent of the University of the Saarland, our team of five presented Mentalytics – When AI knows how you feel– an AI-powered decision support system for personalized rehabilitation and training.

The concept? A real-world public field study where visitors could challenge our AI to predict their perceived exertion during a simple sit-to-stand exercise. Privacy-first, running LLM entirely on local edge computing – no cloud, no data leaving the booth.

The result? Absolutely overwhelming! 🎊

✅ 220 enthusiastic participants took part over the course of the day
✅ People were genuinely excited to "beat the AI"
✅ Every single person completed the questionnaires and squats/sit-to-stand exercise with amazing energy and competitive spirit.
✅ Our locally-hosted Artificial Mental Model (AMM) (accessible via QR code or study tablets) proved that on-edge LLM inference works reliably – even with festival crowds, noise, and diverse devices

The atmosphere was fantastic. Participants ranged from curious families to fitness enthusiasts, all eager to see if the AI could really predict how they felt.

AI in Sports and Beyond 🏃‍♂️

Prof. Wolfgang Maaß also had the pleasure of being interviewed by Dr. Thorsten Klein on "AI in the Fight Against Doping" – a conversation that spanned from early smartwatch app development for athletes in 2011, through collaboration with the World Anti-Doping Agency (WADA), to current research in Alzheimer's and longevity.

All in all, #TDE25 was a resounding success. Days like these remind us why we do what we do: bringing cutting-edge research out of the lab and into real conversations with real people about the future we're building together.

A huge thank you to our incredible team who made this day such a positive, fun-filled success! 🙌

#DFKI #Mentalytics #TagderDeutschenEinheit #TDE25 #ArtificialIntelligence #DigitalHealth #Rehabilitation #SmartServiceEngineering #Saarbrücken #Saarland #FutureThroughChange

Additional Resources

https://www.linkedin.com/posts/dfki_tde-mentalytics-llavaguard-activity-7380986425208397824-kmXt?utm_source=share&utm_medium=member_desktop&rcm=ACoAACNPxUYBrix0N67kIC0GJWOf-DMT2aDf0WE

Share this article

Back to News Overview

Other News

July 30, 2026

ESCADE Closing Event: Three Years of Research on Energy-Efficient AI

On 30 July 2026, more than 50 guests from research, industry and public administration came to DFKI Saarbrücken for the closing event of the ESCADE project. After three years, the consortium coordinated by Prof. Wolfgang Maaß presented how the energy demand of AI can be reduced by up to 90 percent – and drew wide media attention in the days around the event.

July 23, 2026

How AI Becomes Significantly More Energy Efficient – Press Coverage of the ESCADE Results

Ahead of the closing event of the ESCADE project, Saarland University issued a press release on the results of the research consortium led by Prof. Wolfgang Maaß: how the electricity demand of AI can be cut by up to 90 percent through smaller models, automated network design and neuromorphic hardware.

July 11, 2026

Paper Presentation at ICML 2026

Maxx Richard Rahman presented STT-LLM, a structural-temporal tokenization framework for adapting Large Language Models to longitudinal clinical profiles, in the Main Track of ICML 2026 in Seoul.