The scientific method, operationalized
Remyx exists because AI development is an empirical discipline, and the teams doing it deserve tooling that treats it like one.
A decade of experiments behind the thesis
We've spent ten years shipping production ML together, across robotics, healthcare, content recommendation, and enterprise data infrastructure, and several more building open-source vision-language tooling the research community adopted. Across all of it, the same pattern held. Teams move fastest when every experiment is deliberate, every result is recorded, and every decision feeds the next one. Remyx is that practice, productized.
Founders
Salma Mayorquin
ceo & co-founder
Ex-Databricks Solutions Architect, advising MLOps strategy from startups to Fortune 500. Applied Mathematics, UC Berkeley. 8+ years in ML infrastructure. Recognized by NVIDIA's developer community.
Terry Rodriguez
cto & co-founder
Ex-Riot Games, Robust.AI, and Tubi. 10+ years as an ML engineer and data scientist across healthcare, robotics, and content recommendation. UNC Chapel Hill and UC Berkeley. Open-source tools cited by Google DeepMind.
We named the conference after the unit of progress
On October 30, 2025 we hosted the inaugural Experiment, a gathering for practitioners who treat AI development as the empirical discipline it is. Over 900 people registered and 200+ attended for a day of talks on experiment-driven development, evaluation in production, and learning from the teams who ship this way at scale.
The conversations from that room shaped the platform you see on this site, and they convinced us the community around this practice deserves a standing home. Experiment 2026 is in the works.
# talks, panels, and the hallway track from the inaugural event
Built in the open
Our open-source work (VQASynth, SpaceLLaVA, and the Space-family VLMs) is where the Remyx thesis was first tested, and it's been highlighted by Google DeepMind, AG2, and Ray. We write regularly about experiment-driven AI development on our blog and Substack.
The product evolved as we learned
Remyx began by automating model creation. Building and operating those systems taught us that producing another model was rarely the enduring bottleneck. The harder problem was deciding which change was worth evaluating, preserving what the team learned, and carrying that evidence into the next decision.
Today's Remyx productizes that lesson. You may find older material about the earlier product around the web. We keep it up because the through-line matters: every version of Remyx has been about making AI development an evidence-driven practice.
Building AI into your product? Talk to us
We're in early access with a small group of teams shipping AI in production. If your team wants every experiment to count, we'd like to hear from you.