Castane AI highlights four runtime tricks for LLM inference, including continuous batching, page attention, prefix caching, and speculative decoding.

Castane AI is a Paris-based artificial intelligence studio. We build custom AI applications for companies, and we train people to build and use AI themselves. Three ways we work: - Consulting & custom development. Our developers design, build and deploy AI applications tailored to your business — automation, AI agents, document search, data analysis — from initial scoping through to production. Engagements are sized to your organisation, from small businesses taking a first step to large groups deploying at scale. - Training & workshops. Practical workshops adapted to each audience: executive briefings for leadership, applied sessions for managers, and hands-on technical training for developers. Your teams learn from real examples drawn from your own work, and we cover best practices, security and governance. - Bootcamps. Intensive programmes for individuals who want to build. In a few weeks, participants learn to design, develop and deploy a real, working application — and leave with a project in their portfolio. We build AI — and teach you to.

Founded 2022