We design, develop, and evaluate AI systems that run the real world, and shape the researchers and founders who will define its future. Based at ETH Zurich.
Zurich, Switzerland
Joined August 2025
- Tweets78
- Following62
- Followers453
- Likes237
Agentic Systems Lab | ETH Zurich retweeted
From a beer at Stanford to a frontier AI lab in Zurich: ๐ป๐จ๐ญ
I've joined Aionic Labs, alongside my work at ETH's Agentic Systems Lab!
I've known the four founders as friends and colleagues for years. When they reached out about joining the team, it was an easy yes! ๐
โ๐ ๐ฐ๐ถ ๐ฅ๐ฐ๐ฏ'๐ต ๐ฏ๐ฐ๐ณ๐ฎ๐ข๐ญ๐ญ๐บ ๐ญ๐ฆ๐ข๐ณ๐ฏ ๐ฉ๐ฐ๐ธ ๐ต๐ฐ ๐ฃ๐ถ๐ช๐ญ๐ฅ ๐ง๐ฐ๐ถ๐ฏ๐ฅ๐ข๐ต๐ช๐ฐ๐ฏ ๐ฎ๐ฐ๐ฅ๐ฆ๐ญ๐ด ๐ช๐ฏ ๐ข๐ค๐ข๐ฅ๐ฆ๐ฎ๐ช๐ข.โ
In this weekโs lecture on ๐๐ฎ๐ข๐ฅ๐๐ข๐ง๐ ๐๐ ๐๐ง๐ญ๐ข๐ ๐๐ ๐๐ฒ๐ฌ๐ญ๐๐ฆ๐ฌ ๐๐จ๐ซ ๐๐ง๐๐ฎ๐ฌ๐ญ๐ซ๐ฒ ๐๐ซ๐๐ง๐ฌ๐๐จ๐ซ๐ฆ๐๐ญ๐ข๐จ๐ง at ETH Zurich, @nzuma0 and @PatrickLangerAI covered large language model (LLM) training and adaptation.
One fascinating aspects of foundation models we discussed was ๐๐ฆ๐๐ซ๐ ๐๐ง๐ญ ๐๐๐ก๐๐ฏ๐ข๐จ๐ซ. You keep scaling data, model size, and compute, and suddenly models start demonstrating capabilities they weren't explicitly trained to perform. These capabilities are difficult to predict from smaller-scale experiments.
But here's the challenge: ๐๐ง ๐๐๐๐๐๐ฆ๐ข๐, ๐ฒ๐จ๐ฎ ๐ซ๐๐ซ๐๐ฅ๐ฒ ๐ก๐๐ฏ๐ ๐๐๐๐๐ฌ๐ฌ ๐ญ๐จ ๐ญ๐ก๐ ๐๐ง๐ ๐ข๐ง๐๐๐ซ๐ข๐ง๐ ๐๐๐ฉ๐๐๐ข๐ฅ๐ข๐ญ๐ข๐๐ฌ, ๐ข๐ง๐๐ซ๐๐ฌ๐ญ๐ซ๐ฎ๐๐ญ๐ฎ๐ซ๐, ๐๐ง๐ ๐ฌ๐๐๐ฅ๐ ๐ง๐๐๐๐๐ ๐ญ๐จ ๐๐ฑ๐ฉ๐๐ซ๐ข๐๐ง๐๐ ๐ญ๐ก๐๐ฌ๐ ๐ฉ๐ก๐๐ง๐จ๐ฆ๐๐ง๐ ๐๐ข๐ซ๐ฌ๐ญ๐ก๐๐ง๐.
We covered post-training methods for LLMs and explored emergence and scaling laws. These topics help explain how model capabilities develop, how model behavior can be shaped, and what drives recent breakthroughs in LLMs.
For anyone building with AI, understanding these foundations helps inform decisions about what to automate, how to evaluate systems, and where human oversight matters.
Thanks to Nicolas Zumarraga and Patrick Langer for sharing insights from their research and real-world experience at Aionic Labs and Meta.
The Building Agentic AI Systems for Industry Transformation lecture takes place every Wednesday (16:15โ18:00) at ETH Zurich and is led by @kev_osull, @PatrickLangerAI, @robertjakob, and Markus Kreft.
Lecture link in the comments. ๐๐
Lecture Link:
vvz.ethz.ch/Vorlesungsverzeiโฆ
Lecture recordings will be made available here once ready.
Agentic Systems Lab | ETH Zurich retweeted
Was fun sharing the stage with @PascalBtd at AI+X, looking back at the early days of the @ETH_agent_lab and what we're building now with @aionic_labs and @seldon_tech.
Thanks @ETH_AI_Center, Sรถren Fehrmann and Swisscom Ventures for having us!
Agentic Systems Lab | ETH Zurich retweeted
Frontier Time Series Language Models struggle with precise anomaly localization over long contexts, when useful signals might be hidden in hours of recordings.
In our new paper we trained a tiny Recursive Language Model to SoTA performance on long context time series.
We introduce AnomalyXL, an infinitely scalable synthetic time-series generator with programmatic ground truth answers.
Training our RLM formulation on this RL environment transfers to real world anomaly detection in Software Observability, ECG and Sleep data.
๐งต๐
Agentic Systems Lab | ETH Zurich retweeted
Still a student? Want to launch a startup and raise your first angel round? Hereโs the blueprint.
Pascal Bertrand (@PascalBtd) & Atoof Shakir (@atoof_sh) join Europe on Edge to talk about building an AI startup at ETH, raising without a pitch deck, and teaching AI to see.
๐ฅ Trailer below
๐ง Full episode in the first reply
Agentic Systems Lab | ETH Zurich retweeted
The world of R&D is forking into two paths: the token-abundant research, and the token-starved research. The future is in the former - evidentially, the progress by today's top AI industry teams and neolabs is breathtaking, where researchersโ human brilliance is super charged by AIโs assistance. Every research university president should be reading this report and reflecting on what the future of higher education research should be.
openai.com/index/research-acโฆ
Agentic Systems Lab | ETH Zurich retweeted
Introducing VGI-Bench: a multimodal, holistic benchmark probing 12 distinct visual and audio-visual skills.
550 human-curated questions, designed to mitigate the common mistakes in today's video benchmarks and expose pragmatic failures of state-of-the-art models.
Best model: 64.73%. Humans: 84.5%.
Agentic Systems Lab | ETH Zurich retweeted
Today weโre launching browsers for nunu!
Our agents can now browse the web just like humans do, built with @browser_use ๐ค
Weโre also introducing cross-platform support, allowing agents to run automations across PCs, phones and web.
Here are some examples โฌ๏ธ
Agentic Systems Lab | ETH Zurich retweeted
Yesterday, we hosted Curious Machines, our first event with talks from @matteosaponati and @lsteno about their work on auto-research (more about that soon!). Thanks to everyone who attended and for those we could not accommodate this time, stay tuned for our next events
Thanks to @ETH_agent_lab for the location and @GHollbeck for taking pictures
Agentic Systems Lab | ETH Zurich retweeted
We are launching Aionic Labs (@aionic_labs). ๐ฅ๐
With support from SPRIN-Dโs Next Frontier AI initiative, we are building a frontier AI company at the heart of Europeโs leading AI ecosystem.
Clinicians decide when a patient needs intervention. Factory teams catch machine problems before production stops. Operators determine which alarms demand action.
These are some of the most consequential decisions in the physical economy, and they all depend on time-series data.
We are developing Time-Series Language Models (TSLMs) that can understand these signals, reason over them, predict what comes next, and suggest actions in natural language.
TSLMs connect sensor data with language experts use to make critical decisions.
Europe is the right place to build this technology. We have world-class research, top technical talent, and decades of industrial expertise and data.
Our team brings together former founders, frontier tech experience, and world-class researchers from Stanford, Harvard, ETH Agentic Systems Lab (@ETH_agent_lab), ETH AI Center (@ETH_AI_Center) and CDTM (@cdtm_munich). We have built companies, developed technology at scale, and advanced research at the frontier of AI.
Over the coming months, we will:
โ Launch selected TSLM pilots across manufacturing, healthcare, energy, logistics, telecom, insurance, and banking.
โ Build an open research community around OpenTSLM (@OpenTSLM) and TimeNet.
โ Enable trusted data partnerships across industries.
โ Accelerate the Time-Series AI community in Zurich and Europe.
From Zurich. For Europe. ๐จ๐ญ๐ช๐บ
Join our mission and help build a Frontier AI Lab from day one. Career and pilot program links are in the comments.
Managing Directors: Robert Jakob (@robertjakob), Thomas Kaar (@thomaskaar)
Supported by: SPRIN-D (@SPRIND)
Photo: Felix Adler fรผr SPRIND
Earlier this year, we launched the first student batch of the ๐๐ง๐ ๐๐ด๐ฒ๐ป๐๐ถ๐ฐ ๐ฆ๐๐๐๐ฒ๐บ๐ ๐๐ฎ๐ฏ. Today, applications for Batch 2 are officially open! ๐
๐ฆ๐๐๐ฑ๐ฒ๐ป๐๐ ๐ถ๐ป ๐ผ๐๐ฟ ๐ณ๐ถ๐ฟ๐๐ ๐๐ฎ๐๐ฐ๐ต ๐ต๐ฎ๐๐ฒ ๐ฎ๐น๐ฟ๐ฒ๐ฎ๐ฑ๐:
โ Published papers at ICLR and ICML
โ Advised executives at SMI-listed companies
โ Raised seven-figure funding for their startups
โ Secured full-time offers from frontier AI labs
Today, our work spans the cutting edge of AI, from Agentic Systems and Multimodal AI to TSLMs and RLMs โ applied across sectors like healthcare, energy, financial services, manufacturing, and robotics.
๐ก๐ผ๐, ๐ถ๐โ๐ ๐๐ถ๐บ๐ฒ ๐ณ๐ผ๐ฟ ๐๐ต๐ฒ ๐๐ฒ๐ฐ๐ผ๐ป๐ฑ ๐๐ฎ๐๐ฐ๐ต.
The goal? Help you turn your semester project, bachelorโs thesis, or masterโs thesis into frontier AI research, an AI startup, or a transformative industry collaboration. With you in the driver's seat.
๐๐ฝ๐ฝ๐น๐ถ๐ฐ๐ฎ๐๐ถ๐ผ๐ป๐ ๐ฎ๐ฟ๐ฒ ๐ผ๐ฝ๐ฒ๐ป ๐๐ผ ๐๐ง๐ ๐๐๐๐ฑ๐ฒ๐ป๐๐ ๐ฎ๐ป๐ฑ ๐ด๐๐ฒ๐๐ ๐๐๐๐ฑ๐ฒ๐ป๐๐ ๐ณ๐ฟ๐ผ๐บ ๐ผ๐๐ต๐ฒ๐ฟ ๐๐ป๐ถ๐๐ฒ๐ฟ๐๐ถ๐๐ถ๐ฒ๐.
Who's next? ๐
๐ Link to apply in the comments.
You can find more information about the lab, available student tracks, and the application process here:
๐ย Lab Website: agenticsystemslab.org/
๐ย Student Tracks: agenticsystemslab.org/#opporโฆ
๐ย Application Form: docs.google.com/forms/d/e/1Fโฆ
Excited to share that members of our lab co-authored 6 papers accepted at #ICML2026, including three Main Track and three Workshop papers ๐ฅ๐
๐ Accepted papers:
โช๏ธ OpenTSLM: Time-Series Language Models for Reasoning over Multivariate Medical Text- and Time-Series Data [Main Track]
โช๏ธ Auditing Emotion-Vector-Steered Political Bias in Open-Weight LLMs [AI4GOOD Workshop]
โช๏ธ Reinforcement Learning of Karma Bidding Strategies [NExT-Game Workshop]
โช๏ธ Cinematic Source Separation with Dialogue-Driven Sidechain Ducking [Workshop on Machine Learning for Audio]
โช๏ธ Reinforcement Learning for Tool-Calling Agents in Fast Healthcare Interoperability Resources (FHIR) [Main Track]
โช๏ธ Preserving Plasticity in Continual Learning via Dynamical Isometry [Main Track]
@robertjakob @PatrickLanger20 @gaborhollbeck @f14wn @DerRiehl @atoof_sh @deepqlearning @kev_osull @cs06thegreat @nzuma0 @maxrosenblattl
Huge congratulations to everyone involved. We are looking forward to presenting these works, reconnecting with colleagues, and meeting new friends in Seoul ๐ฐ๐ท
๐ Full links in the comments.
๐ Paper 6/6
Preserving Plasticity in Continual Learning via Dynamical Isometry
Rosseau A, Mรผller R, Nowe A.
Accepted at ICML 2026 Main Track
ICML: icml.cc/virtual/2026/poster/โฆ
@deepqlearning
๐ Paper 5/6โ
Reinforcement Learning for Tool-Calling Agents in Fast Healthcare Interoperability Resources (FHIR)
Knorr M*, Mรผller R*, Bremer JP, Schweingruber N.
Accepted at ICML 2026 Main Track
arXiv: arxiv.org/pdf/2605.14126
ICML: icml.cc/virtual/2026/poster/โฆ
@deepqlearning
๐ Paper 4/6
Cinematic Source Separation with Dialogue-Driven Sidechain Ducking
Shakir A, Groetschla F, Lanzendoerfer L, Wattenhofer R.
Accepted at ICML 2026 Workshop on Machine Learning for Audio
Workshop: mlforaudioworkshop.github.ioโฆ
@atoof_sh
๐ Paper 3/6
Multi-Agent Reinforcement Learning of Karma Bidding Strategies
Riehl K, Psarou A, Mรผller R, Wu F, Langer P, Jakob R, Hollbeck G, Kouvelas A, Kucharski R, Makridis MA.
Accepted at ICML 2026 NExT-Game Workshop
OpenReview: openreview.net/forum?id=HaUUโฆ
@DerRiehl @deepqlearning @PatrickLanger20 @robertjakob @gaborhollbeck
๐ Paper 2/6
Auditing Emotion-Vector-Steered Political Bias in Open-Weight LLMs
Hollbeck G, Peters B, von Recum A, Riehl K, McGail A, Windeck J, OโSullivan K, Jakob R.
Accepted at ICML 2026 Workshop: AI4GOOD
OpenReview: openreview.net/forum?id=v3dsโฆ
@gaborhollbeck @f14wn @DerRiehl @kev_osull @robertjakob
๐ Paper 1/6
OpenTSLM: Time-Series Language Models for Reasoning over Multivariate Medical Text- and Time-Series Data
Langer P, Kaar T, Rosenblattl M, Xu MA, Chow W, Maritsch M, Jakob R, Wang N, Liu J, Verma A, Han B, Kim DS, Chubb H, Ceresnak S, Zahedivash A, Sandhu ATS, Rodriguez F, McDuff D, Fleisch E, Aalami O, Barata F, Schmiedmayer P.
Accepted at ICML 2026 Main Track
arXiv: arxiv.org/abs/2510.02410
Github: github.com/StanfordBDHG/Openโฆ (+1200 โญ๏ธ)
ICML: icml.cc/virtual/2026/poster/โฆ
Website: opentslm.com/
X: @OpenTSLM
@PatrickLanger20 @robertjakob @cs06thegreat @nzuma0 @maxrosenblattl @maxxu05 @PSchmiedmayer
