

What It Takes to Build a Frontier Local Coding Model : A conversation with Ricursive Intelligence and Together AI
Join Dan Zhang from Ricursive Intelligence and Zain Hasan from Together AI for a technical session on how the Ricursive Intelligence team built Ripple-65B Preview.
Ricursive built Ripple-65B Preview to begin closing the loop between models and the hardware they run on. An internal variant of the model, trained from scratch, powers Ricursive’s AI chip design platform. Ripple-65B Preview is the general-purpose variant: a dense, 65B-parameter Transformer pretrained on 12 trillion tokens, with support for up to 1M context. Its small parameter footprint lets it run on local devices like NVIDIA DGX Spark.
Ricursive’s launch post covers much of the build. This session goes behind the build: the experiments and ablations, why each decision was made, what didn’t work, and the trade-offs along the way. If you’re an ML engineer or researcher working on pre-training, post-training or RL, or a developer who wants to understand how the models behind coding agents are architected, trained and evaluated, this session is for you.
What we’ll cover
Experiments & ablations, model architecture and training data strategy for Ripple-65B Preview
A look inside actual RL runs, including what worked and what Ricursive learnt
The engineering decisions, unexpected findings, and practical lessons that shaped the development of Ripple.
How to run Ripple on Together and integrate it into coding agents and developer workflows
Register to join the webinar live, ask questions, and receive a recording of the talk.
⚠️ Please provide your valid corporate email upon registration.