While traditional demo days often feel like exercises in pitch perfection and networking optimization, last Thursday in San Francisco, we experienced something different: a gathering of builders who see possibilities others don’t and feel compelled to bring them to life, line of code by line of code.
What began with presentations transformed into an actual science fair, where guests explored each technology hands-on and engaged in deep conversations with their creators.
Typical demo day formalities melted away. Instead of rushed networking, people clustered around demos, asking questions, sharing ideas, and discovering natural collaborations. For a few short hours, we glimpsed a future built on a foundation of shared purpose and collaborative spirit.
Training, Fine Tuning & Models
Anastasia and Pierre-Carl of Pleias announced a suite of models trained entirely on open data. Pleias 1.0, including Pleias-3b, Pleias-1b, and Pleias-350m, democratizes AI knowledge at a time when some of the planet’s most widely used models remain tucked away in walled gardens.
Michael of Sartify spoke about PAWA, a chatbot built on their Swahili-language models. Sartify’s work isn’t just about localization — it’s about ensuring AI speaks everyone’s language, literally and figuratively. Meanwhile, Gemma Turon of Ersilia told us how they are bringing AI-powered drug discovery tools to scientists studying infectious diseases in the Global South.
As Miquel Duran-Frigola, Ersilia’s Chief Scientific Officer and Co-Founder, notes in his piece, Beyond the Kingdom of the Sick, “Most AI narratives, and the aesthetic and vocabulary surrounding them, are marked by the allure of acceleration, augmentation and, ultimately, the promise of liberation from our own physical and intellectual boundaries. However, the value of AI may lie less in futuristic enhancement than in addressing fundamental healthcare needs.”
Tony and Ali from Transformer Lab are building a dead simple and highly performant toolkit for working with LLMs on your local machine. And Koel Labs showcased its real-time pronunciation feedback tool.
Developer Tools & Infrastructure
Jonas flew in from Germany to showcase how Theia IDE can pair with local AI solutions like Mozilla’s starcoder2-llamafile to transform programming workflows. Open WebUI proved that building sophisticated AI applications doesn’t require corporate backing — just vision and technical rigor. (Tim is a team of one btw. A team of one with 3M+ downloads and 50k+ GitHub stars.)
Want to know what’s happening under the hood of an audio model? Talk to Konstantine of FREUD or read the analysis that inspired his talk.
Jeremy Lewi pulled out the stops for his Foyle presentation, showcasing how his copilot enhances his day-to-day work and will transform DevOps. (Hint: it’s about predicting commands given the cells preceding them.) And Marker is sure to add a few more stars to its 18.5k with the latest table formatting optimizations and general speedup.
Is the future of AI training distributed? By bringing the power of numerical computing to Elixir, NX offers one of the most fault-tolerant and performant solutions.
Media & Data Visualization
Vishnu and Laurens flew from India to show off Ente, their open-source end-to-end encrypted alternative to Google and Apple Photos. If you’re privacy-minded and not convinced you need to switch, check out their experiment, “They See Your Photos,” written up in Wired.
Some projects pushed even further. Latent Scope reimagined how we visualize unstructured data, while Tölvera blurred the lines between art and artificial life, creating simulations that challenge our very understanding of intelligence.
The next chapter of computing will be written by those who choose to build rather than acquiesce. By those who ask not just “Can we build this?” but “Should we build this?” By those who understand that the most powerful technology is that which makes us more human, not less.
