Smaller Models, Sharper Tools: Building with Machine Learning to Drive Product

​Most AI tooling today defaults to "just add an LLM." At Cinder, we've been taking a different path: using Machine Learning models to help non-technical stakeholders rapidly build and iterate on systems without turning every workflow into a LLM prompt-engineering exercise.

​This talk will cover how we built smaller, targeted multimodal models to deliver high-quality content violation detection at a fraction of the cost and latency of larger generative systems, while still being practical for non-ML specialists to use.

​We'll also share how this foundation is evolving into an agentic training loop for faster evaluation, feedback, and deployment. It's a behind-the-scenes look at applied ML as infrastructure: less demo magic, more scalable product advantage.

A moderated panel featuring Tullie Murrell and Mitch Krieger will run right after the tech talk, covering the different approaches being taken by companies building in this space.