Web AI isn’t just “call an LLM API.” In this talk, you’ll explore a practical toolkit for building AI features that run directly in the browser and Node.js—without per-request costs and with privacy-first defaults.

You’ll go beyond chat and cover Genetic Algorithms (optimization), Reinforcement Learning (decision-making agents), and Transformers.js as a runtime for running transformer-based models in JavaScript for vision, embeddings, classification, and more.

You’ll also see how to add a local knowledge layer using SQLite compiled to WebAssembly to enable RAG-style experiences fully on-device.

Outline

- Web AI without vendor lock-in: what’s possible today
- Genetic Algorithms: fast optimization and creative search
- Reinforcement Learning: agents that learn by trial and reward
- Transformers.js: running transformer models in browser/Node (not just chat)
- Local RAG: embeddings + retrieval + SQLite/WASM
- Shipping to mobile/desktop: Workers, performance, and UX patterns