The Swiss AI SME Circle was launched by the National AI Initiative (SNAI) in collaboration with the National Supercomputing Centre (CSCS) and swissICT. The goal is to bring together Small and Medium Enterprise (SMEs) with researchers and innovators in AI to promote peer learning, knowledge sharing and practical collaboration.

This post shares some thoughts from taking part in the first and second editions, at the AI+X Summit in October and the swissICT HQ in Altstetten in December 2025. These meetings are based on a mandate of widening the scope of industry transfer from research institutions. As an independent consultant working primarily outside of the corporate sector for the past ten years, I felt well within the target audience, and indeed recognized a few of the people around me in the audience.

The Circles were attended by around 50 people both times, starting with a welcome from our hosts and updates from the ETH AI Centre staff that are in charge of the project. Both times we got to hear short inputs from the frontiers of AI development, including by people working on the Apertus project. This is followed by practical reports on real use cases, and in the case of the last event, a peer-learning session in small groups.

1st Circle

The Circles are being co-led by Alicia Rieckhoff and Hanna Brahme, with the events presented as an excellent chance to hear first hand technical experiences relevant to implementations of new AI models, and to discuss ideas for the roadmap of flagship projects like Apertus. This was the meeting where I first met Imanol from the core team.

We also heard from the head of the unit of Research Infrastructure Engineering at CSCS that was involved in training Apertus. Joost shared insights on the sustainability question, comparing the training of AI to other forms of public utilities, e.g. the power envelope of the full ALPS data centre equivalent to two locomotives of the SBB. Also highlighted in the first Circle meeting, was the interest to develop industrial relationships and R&D projects (cscs2go) within the supercomputing facility.

See also my earlier blog post about the topic if you need an intro:

As mentioned above, a key goal is to share use cases from the local community – so in the first SME Circle we heard from Christian Stocker, a well known Software Developer and Co-Founder of the digital agency L//P, who presented his work on ZüriCityGPT: a chat that knows "almost everything that is published on stadt-zuerich.ch" (the official website of the city of Zürich).

This quickly became a widely cited example of a public government AI app in production, and in October it was expanded with support for Apertus in the ZüriCityGPT OSS Version, that you can read about in detail in the L//P blog. Christian also demoed the new StrbGPT version of the bot, a RAG of all the public City Council resolutions since 2010, and continuously updated. By the way, you can also use Apertus in the standard chatbots with a model:apertus prompt parameter.

Christian's team took the time to deal with many challenges in building these prototypes, and mentioned a couple of issues with Apertus that were worked out in the discussions that ensued. A range of suggestions for future development, thoughts on performance and reliability, were expressed in the lively audience discussion that followed.

2nd Circle

The second event was hosted together with Carol Lechner and team at swissICT, who describe themselves as the "largest and most important" network of IT companies in Switzerland, going all the way back to 1955. Based in a rather modest office on the outskirts of Zürich's district 9, next to modern skyscrapers housing leading tech firms, overlooking one of the most important nodal points of the railroad system. There is a calm sense of purpose here that is perhaps lacking in the more urgency-filled offices of avant-garde tech groups.

We were welcomed to the event by Christian Hunziker, Executive Director of swissICT, who briefly told us about the organization, stating that AI is a cross-functional topic addressed by various groups, that is now converging with the working group supporting today's event. Rebecca Brauchli and Sunnie Groeneveld were here to support Hanna and Alicia, and shared some trending topics among swissICT members.

Legal and regulatory matters in particular seem to be of recurrent interest within the group. A quick show of hands of the level of progress. I got the feeling that we were lacking the presence of more resource-constrained SMEs, nevertheless most people seemed to still working to integrate AI into processes: not yet having turned out a revenue stream.

As today's SME use case, we heard a presentation by Monitoris AG - a startup developing AI-based backoffice solutions. The chief executive, Islam Alijaj is also a member of the National Council, and a widely acknowledged disability activist. The opening remarks together with Hermann Arnold inspired me to immediately pay attention and test the app myself. It was possible to quickly generate invoices from text, file upload, or a photo of an existing invoice.

This is a clear practical use case that the nascent community can support. Even though they currently base the product on a big commercial service, they are targetting SME customers who will prioritize high quality and reliability, and were open to discussing legal and engineering topics. For instance, the current accounting export to Bexio will be complemented by Odoo, an open source alternative.

Following this, fellow Swiss {ai} Weeks initiatior Sabine Wildemann announced the new aiLights program: an online network with a focus on academia, private sector and civil society. There are quite clear overlaps, yet distinctly grassroots beginnings here that will hopefully prove synergistic with the efforts of SNAI, swissICT. It felt well received, as a strong complement to today's and upcoming AI Circle sessions.

Sabine and I have been collaborating for months: looking forward to engage with the first batch of "lights" to disuss the Swiss Public Inference Utility in January. Stay tuned! ✨

Peer learning

After these presentations, we swiftly dived into the group discussion. The Circle participants were asked to choose between 4 themes, then to literally stand in a circle and discuss this in detail during a brief peer-learning sessions. The questions aimed at an exchange of best practices, technology and infrastructure, partnerships, culture change, etc.

One person from each circle was asked to briefly present the main inputs. There were plenty of ideas for the kind of information and support that we could except from future activities: several practical use cases, needs in technical support or networking area, regulatory advice and funding opportunities that go in someways beyond the main focus of current industrial networks. The insights we were asked to share were: 1 use case, 1 hurdle, 1 solution hack.

I wandered over to the 2nd group, introducing myself as someone working on research topics in social integration, maintaining data tools, and building community around open legal tech. Standing next to me was a ZHAW researcher who explored using AI to analyze global supply chain risks with satellite data. We talked about a few other ideas, deciding as a group to focus the debate on using open AI models in helping to ensure that such uses do not expose sensitive information or violate regulations.

The main hurdle was the lack of clarity on surveillance capabilities and data protection regulations in commercial settings. Our solution hack proposed to leverage Apertus to create a secure, collaborative platform where diverse stakeholders can input and compare scenarios. We could use privacy-preserving AI models to analyze risks without compromising sensitive data. I fed a summary of these points into Apertus on the Public AI inference utility, and later tidied the resulting discussion in this notepad:

Over coffee and bagels, I had the chance to talk to some very professional and interested people, e.g. working on international governance - in particular concerning the rights of data workers, innovating legal decision workflows, or building infrastructure to monitor and keep in check the proliferation of AI tools everywhere. We talked about Apertus capabilities, exchanged tips on model configuration and prompt writing. This will be a good venue to propose and vet constructively critical ideas in openly developed AI.

Final prompts

During a chat with Apertus itself, a couple of ideas surfaced as well: issues of data protection, collaboration with research, long-term sustainability. Open questions for discussion: avoiding dependency on large tech corporations, ethical considerations in AI adoption, preparing employees for AI changes, sustainability of AI projects in SMEs. I took home some additional 'tips' for future participations: bring a real challenge, note some key discussion topics, explore potential partnerships, get ready to ask questions and be open to new perspectives.

Thanks very much to the AI Centre & swissICT for organizing, hosting us and sharing a delicious lunch: captured above in MyFoodRepo AI. A discussion for another time, which you can find already in bits and pieces (no pun intended) on my social feed.

Feel free to @respond, let me know if you have any questions or ideas for this community - one that I am hoping to support with my full attention in 2026.