Introducing dotAI 2026's Full Speaker Lineup on Sept 17 in Paris: -Ziv Ilan - AI Labs at @nvidia -@spolu - Co-Founder at @DustHQ -Marta Garnelo - Chief Science Officer at @Fundamental -@dlouapre - AI Scientist & Educator at @huggingface -Ian Massingham - Head of Applied AI, EMEA Startups at @AnthropicAI -Pierre-Edouard Lieb - AI Deployment Manager at @OpenAI -Patrick Brosset - Developer Relations PM at @MicrosoftEdge -Gaëtan Brison - AI Manager at @doctolib -Pierre-Louis Cedoz - Head of Research at @hcompany_ai -Guillaume Vernade - Senior Developer Advocate at @GoogleDeepMind Lightning talks speakers: -Guillaume Blaquiere - Group Data Architect at @CarrefourFrance -@gregqualls - Head of PM & Content at @upsundotcom, Co-Builder of askcodex.ai -Léo Arsenin - Solutions Engineer at @Cloudflare -Hélène Philippe - Machine Learning Researcher at @raidium_med -@ngxson - Software Engineer at @huggingface -Aygalic Jara - PhD researcher at Université Paris-Saclay × SCIAM Talk titles dropping soon, follow along for the full schedule! Grab your ticket now dotai.io/

What if your local LLM could answer faster without changing the model you’re running? In “llama.cpp - Faster and Easier for Everyone!” Xuan-Son Nguyen takes us inside speculative decoding: how llama.cpp drafts tokens to speed up inference, using methods like draft models, n-grams, MTP and DFlash / DSpark. The talk wraps up with a live demo - and a look at how quickly you can get started. @ngxson is a software engineer at @HuggingFace, a core maintainer of llama.cpp and a contributor to open-source LLM tooling and on-device AI. 🟡 Replay here: youtube.com/watch?v=bq1cSrr0…

What if an AI could use a website without guessing where to click? In “WebMCP: Stop Scraping the Web, Start Calling It,” Patrick Brosset explores a more reliable way for AI agents to get things done online. Instead of interpreting pages and guessing at clicks, agents can use explicit tools - with structured calls that developers and users can inspect. The talk looks at how this approach could make web workflows less brittle, even as interfaces change. Patrick is a Developer Relations PM at @MicrosoftEdge. He’s spent more than two decades building web apps and working to improve interoperability, developer experience and tooling across Firefox and Chromium. 🟡 Enjoy the replay: youtube.com/watch?v=3NDly38D…

As AI agents write more code, how will developers decide what to trust? In “Will the Future of Code Be Formal or Informal?”, Stanislas Polu looks at a new bottleneck for software teams: reviewing agent-generated code. He explores whether agents could make formal methods practical, and whether developers might one day move beyond code altogether. His argument: the future probably lies between prose and proof, with code supported by focused, structured natural-language contracts that people can review and agents can check. @spolu is co-founder of @DustHQ, which helps enterprises build AI agents using internal data and tools. Before that, he worked on reasoning and automated theorem proving at OpenAI and co-founded TOTEM, later acquired by Stripe. Watch: youtube.com/watch?v=eg8u-010…

When the question is “What happens next?”, maybe don’t ask a language model to write you an essay about it. In “AI for Prediction: Why You Need a Statistician, Not a Philosopher,” Kevin Scaman explores why language models and tabular models excel at different things. Through simple experiments, he looks at where LLMs fall short on prediction - and how models trained on structured data can help with decisions like forecasting demand, spotting risk or predicting customer churn. Kevin is an ML researcher at @Fundamental, where he works on tabular foundation models. Before that, he was a research scientist at Inria Paris, an Associate Professor at École Polytechnique and led an optimization-for-ML team at Huawei’s Paris lab. He also received a NeurIPS Best Paper Award in 2018 for his work on distributed optimization. 🟡 Replay: youtube.com/watch?v=EbUa-LSx…

Prompting got you this far. Now what? In “When Prompting Runs Out: Post-Training an Open Model,” Ziv Ilan lays out how to tell when prompt engineering, RAG and better tools have hit their limits -and what to try next. Using Nemotron as an example, he walks through post-training approaches like SFT, DPO, reinforcement learning and distillation, with public weights, datasets and training recipes to explore yourself. Ziv works with AI labs at @NVIDIA, helping research teams build and scale the systems behind their models. With experience across software engineering, electrical engineering and computer science, he brings a practical infrastructure perspective to the path from AI research to production. Catch the replay: youtube.com/watch?v=poJ1bdZJ…

AI that runs in your browser, on your hardware, without calling the cloud? That’s not boring. That’s pretty handy. In “Boring but Effective: WebAI That Ships,” Nico Martin shows how small open models can handle useful tasks right in the browser - from semantic search to speech recognition and background removal. No API key, no cloud subscription, no giant download. Just WebAI that’s ready to get to work. @nicodotdev is a Machine Learning Engineer focused on WebML at Hugging Face, and a Google Developer Expert in AI and Web Technologies. He makes it easier for web developers to bring machine learning into the browser - and shares what he learns through workshops, articles and talks. 🟡 Catch the replay: youtube.com/watch?v=WfojGcUb…

We thought AI agents would do the busywork. So why are we the ones on call? In “Don’t Become Your Agent’s B****", Guillaume Vernade gets into the messy reality of building with AI agents - and how to design workflows that keep humans in control, without turning us into terminal janitors for hallucinating interns. Guillaume is a Senior Developer Advocate at Google DeepMind. His career has taken him from prototyping for Google Glass and early Android watches to product ownership, Agile coaching and even game production. That’s a lot of perspectives to bring to the question: how do we build agents that work for us, not the other way around? 🟡 Catch his talk here: youtube.com/watch?v=-UHcQD4p… Like what you see? Our next edition of dotAI is October 21 2027 and early bird tickets are available on our website!

Kicking off our dotJS rewind with Jad Joubran - and a look at the web’s glow-up! The web platform is moving quickly: features that once took years to reach browsers are becoming usable sooner, while AI is making it easier to explore APIs and build the first version of a feature. @JoubranJad looks at what’s now possible -and how developers’ work changes as the mechanics of building get easier. Jad is an independent web performance consultant and educator, and the creator of LearnJavaScript.online. He’s taught more than 20,000 developers and was recognized by Google as a Developer Expert. 🟡 Watch his talk: youtube.com/watch?v=tH5JDK5F…

First stop in our dotAI rewind: David Louapre’s talk on opening up the black box of AI. How can we look inside a neural network and understand what it’s doing? @dlouapre introduces research into how language and vision models represent concepts - and how those representations can be identified and influenced. He also looks at what this work can tell us about model behavior and safety, including when working with open-source models. David is an AI scientist and educator at @HuggingFace, and the creator of Science Étonnante - a science YouTube channel with more than 1.5 million subscribers, making complex ideas wonderfully easy to get curious about. Watch the replay here: youtube.com/watch?v=G85qi0_1…

Feels like dotAI and dotJS were just yesterday! In reality, it’s been two weeks, so here’s a quick refresher: the replays are now online 🎉 . If you couldn’t make it, or want to revisit a talk, enjoy the replays: 🟡 dotAI: youtube.com/playlist?list=PL… 🟡 dotJS: youtube.com/playlist?list=PL… On stage with amazing speakers from different backgrounds and companies such as @NVIDIA, @AnthropicAI, @Meta, @HuggingFace, @upsundotcom, @Cloudflare, @CarrefourFrance, @Mistral, @Microsoft, @Doctolib, Mirakl, @GoogleDeepMind, @Fundamental, @HCompany_ai, VIRGIO, SCIAM, @Alpic_ai, @GitHub, @Raidium_med, @DustHQ and more!

Yesterday was dotAI, today it's dotJS - and what a way to kick off our two days of conferences. 🎉 Huge thank you to everyone who made yesterday unforgettable: our incredible speakers who shared their time and expertise, the volunteers who kept everything running behind the scenes, and our partners who made it all possible. And if you were in the room, you probably noticed someone quietly capturing the day in a whole different way - our illustrator Florine Quincampoix was live-sketching throughout the event, and we cannot wait to share the full/final collection soon. Here's a little sneak peek 👀

Two @Mistral minds, one big idea - next up for dotAI: Joffrey Thomas & Henry Langarde!

dotAI is in just two days! Say hello to our next dotAI speaker: Christian Keller from @Meta! Christian will be speaking about: Beyond the Chatbox: What Changes When AI Runs Where You Are. Most AI is built as if the answer is always a language model in a data center, reached through a text box. That breaks the moment your product runs on a laptop, a phone, or a pair of glasses. Christian starts from the constraint instead of the model: how form factor drives model choice, why specialized models solve problems no amount of prompting will, and why open weights make that design space navigable. Christian is Product Lead for AI Research at Meta Superintelligence Labs, working on generative AI, world models, and reasoning. Before Llama 3, Llama 4, Muse, and AI agents, he led product for PyTorch, launching PyTorch 2.0 and ExecuTorch, bringing AI at scale to the open source community and to billions of people through deployments across wearables and Meta apps. A three-time startup founder and investor, he holds an MBA from Stanford and MS degrees from Georgia Tech and CentraleSupélec. The chatbox was just the beginning, catch Christian live at dotAI to see what's next. 🟡 dotai.io

Same models, wildly different results. Next up for dotAI: Ian Massingham! Ian will be speaking about: How Anthropic Uses Claude Every developer in the room can use the same models @AnthropicAI does. So why do some teams turn them into measured results while others are still running demos? Ian walks through the five operating principles behind how radical new approaches to working with Claude have driven productivity at Anthropic, complete with engineering, security, and sales cases you can copy straight into your own organization. Ian leads Applied AI for startups across EMEA at Anthropic, where he and his team help founders and startup engineering teams build production AI products on Claude. He brings two decades of experience spanning engineering, developer relations, and go-to-market leadership. Previously, Ian was an early team member at Griptape, a startup focused on generative AI in visual effects, and served as CMO at Aiven, the Finnish open-source data infrastructure unicorn. Before that, he spent several years at AWS leading developer relations and solutions architecture, helping hundreds of thousands of developers adopt cloud and AI technologies. Straight from the source, straight to the stage, catch Ian live at dotAI. 🟡 dotai.io

When "good enough" isn't good enough, especially in healthcare. Next up for dotAI: Gaëtan Brison! Gaëtan will be speaking about: The Moving Target: Benchmarking Dynamic AI Systems End-to-End AI systems don't stand still, so why should our evaluation frameworks? At @Doctolib, Gaëtan and his team are building a complex ecosystem of AI agents spanning triage, clinical interview, retrieval, memory, booking, and image understanding, and evaluating each component in isolation is no longer enough. He introduces the concept of synthetic users: AI-powered personas that simulate realistic patient journeys, enabling systematic evaluation at scale without compromising real patient data. He'll also dig into groundedness: how well agent responses stay anchored to verified sources and real patient context, rather than drifting into confabulation across long multi-turn conversations. Gaëtan is an AI Manager at Doctolib, where he works closely with the AI Lab to enhance models through retrieval and post-training approaches, and has contributed to several open-source initiatives in NLP, computer vision, and optimization. Previously, he served as an AI Research Engineering Manager at Institut Polytechnique de Paris and also worked at NYU and Amazon, with a focus on foundation models and multimodal systems. In healthcare, the stakes don't allow for guesswork, catch Gaëtan live next week. 🟡 dotai.io

No ground truth? No problem. Meet our next dotAI speaker from Mirakl: Diane Le Douaron ! Diane will be speaking about: How to Be Correct at Scale With No Ground Truth. Most LLM work can tolerate a near miss. Catalog transformation cannot: expected outputs like fixed value lists and fixed rules demand a deterministic standard of correctness. Diane digs into how you evaluate correctness when capabilities span several pipeline blocks and there's no ground truth to check against, a problem that gets a lot more interesting once "pretty close" isn't good enough. Diane is Staff AI Product Manager at Mirakl, where she leads the AI Catalog product suite as well as new product initiatives, combining a technical and business background to build scalable AI products. She previously held senior product roles at two UK unicorns and ran her own AI strategy consultancy, specializing in Decision Intelligence. Some talks give you answers. This one gives you a whole new way to ask the question! Join us 👉 dotai.io

What has our partner @Upsundotcom (formerly Platform.sh) been up to? 👀 Most engineering teams are running AI agents today. Very few can govern, trace, or predict the cost of what those agents are doing. Upsun Dispatch™ is built to close that gap: governed agent workflows, isolated sandboxes, cross-functional human gates, and cost transparency by default. Launching 17 September at dotAI 2026, don't miss out. Try it for free: lnkd.in/eU4KANWi

Hello Stanislas Polu who will be joining us on stage at dotAI! @spolu will be speaking about: Will the future of code be formal or informal? Human attention, in the form of reviewing agent-generated code, has become the bottleneck for modern software teams. As code gets cheaper to write and more expensive to trust, will it stay the primary artifact developers work with? Stanislas explores the spectrum from formal methods made practical by agents to futures where developers no longer touch code at all, and argues that the interesting path forward lies between prose and proof: code augmented by local, structured natural-language contracts that humans can review and agents can audit. Co-founder of @DustHQ, a platform enabling enterprises to build and deploy AI agents powered by internal data (Salesforce, GitHub, GSuite) and actions (MCP). He previously co-founded TOTEM in 2011, a data analytics company acquired by Stripe in 2014, then joined OpenAI as a research engineer (2019-2022) on Greg Brockman's team, co-authoring papers with Ilya Sutskever on formal mathematics and automated theorem proving. Formal, informal, or something in between? Find out soon... Join us with your team: 🟡 dotai.io

Building AI systems is exciting - until something breaks and nobody knows why... This is where our partner @Datadoghq comes in! Second shoutout, same excitement. Datadog is the leading observability and security platform for the AI era, providing businesses with unified visibility across the technology stack to manage complexity at scale. It brings applications, infrastructure, data, models, and security into one place, using AI to detect and resolve issues before they impact customers. Trusted globally by Fortune 500 companies and high-growth AI leaders, Datadog enables businesses to move faster with clarity and confidence. Join us to meet the team: 🟡 dotai.io (Sept 17)

dotAI is coming up fast, say hello to: Hélène PHILIPPE! Hélène will be speaking about: Radiological AGI: How Close Are We? AI in radiology has split into two camps: specialized tools that excel at narrow tasks but don't generalize, and vision-language models whose versatility is undermined by hallucinations, imprecise measurements, and poor reproducibility. Hélène's work looks at how agentic systems could bridge that gap, pointing toward autonomous radiological assistants and what real progress on Radiological AGI actually looks like. Hélène is a Machine Learning Engineer and PhD student at @Raidium, a startup building foundation models for radiology. Her work is translational, taking these models out of the lab and into clinical practice alongside radiologists. Come and join is to hear this live! 🟡 dotai.io