Most of use reach for a frontier model by default and pay for it on every call, in latency, in energy, in cash, and in everything that leaves their stack. For most of those calls, a small local model would do the job.

RL Nabors, former Meta/React core team member and AWS alum, covers the vocabulary you need to reason about model performance (capability evals, golden datasets, LLM-as-judge) and walks through real cases: a local agentic harness replacing a frontier call, an in-browser moderation classifier defended with production-trace evals, and a generative summarization feature where the rubric turns out to be harder than the model. You'll leave with a framework for deciding when to choose large and off-prem or small and local models, and how to measure your way to the answer instead of guessing.

You will learn:

The vocabulary to reason about model performance (capability evals, golden datasets, LLM-as-judge). A framework for deciding when a small or local model can replace a frontier one and when it can't. A repeatable process for building capability evals from your own production traces, not someone else's benchmark. Working examples of using eval results to iterate on prompts and ship with confidence instead of vibes.

Zero is a new library for Local-first development that automatically replicates a subset of your Postgres database in your browser.

Zero is an excellent fit for Svelte. If I had to say in one line why Svelte feels magical, it's that there is no trade-off between developer experience and user experience - it makes it easier to write better software. Using Svelte, some problems just disappear entirely, and we are given tools to make other problems much easier to tackle. Zero extends this magic. Delays waiting for page loaders and form submissions disappear, and pages update instantly in response to changes made by other users. Zero is currently in Alpha but already in a state where it's quicker to build a feature with Zero than with regular Svelte loaders and actions.

In this talk I'll go over how to set up Zero, what it can do for you, how it compares to alternatives, why it's such a good fit for Svelte, and why I think that this is the first usable product in a new category that will change the way we write a lot of software.

React frameworks like Next.js and Remix restrict and dictate your stack in ways they shouldn't.

Vike is a (React) framework designed from the ground up to be modular:

Choose your architecture: SPA/SSR/SSG, RPC/GraphQL/REST, server/static/edge deployment, ... Use tools with zero-config integration (Vike extensions) or fully control integration code.

Vike embraces the do-one-thing-do-it-well philosophy for unprecedented flexibility and stability. For example, if you have an existing backend with a GraphQL API, then you don't need Server Actions. With Vike, you can use React your way — such as first-class support for GraphQL instead of Server Actions.

Your stack, your choice.

Le Chat is the app developed by Mistral AI, the pioneering French artificial intelligence company. Like ChatGPT or DeepSeek, Le Chat allows everyone, everywhere, to chat with Mistral’s models. Published a few months ago, the app is powered by React Native and Expo.

The talk will be a technical dive into one central building block of the app: text streaming. Allowing LLM’s answers to appear word by word on screen,  HTTP streaming is a classic method - yet a polyfill is still required for React Native's fetch. Why? How does it work exactly? What are the specific native considerations around it? And what else does it power within the React Native ecosystem?