Reliable AI at Scale: Lessons from Dataiku and Mistral AI

Two talks on what breaks when LLM agents go into production, at Dataiku's Paris office.

From 18:30, Maxime Appé, Technical Product Manager, and Thierry Gengler, Engineering Manager, both at Dataiku, on closing the agent reliability gap: agents that drift, loop, lose context, ignore constraints or add latency, and how structured visual agents put control back around them. From 19:15, Harizo Rajaona, Developer Relations Expert at Mistral AI, on serving a text-to-speech model at scale, covering tokenising audio without losing information, pairing a decoder with flow-matching transformers and streaming in real time, with Mistral's Voxtral TTS as the case.

Open discussion and drinks after, to 21:30.

Why go: agent reliability is the problem everyone running agents has and almost nobody writes up straight. Two teams shipping it will. Best for AI and platform engineers already in production.