Join us at Experiment 2025The bottleneck isn't building.
The bottleneck isn't building.
It's knowing what works.
Remyx gives your team a systematic way to discover what's relevant, test what's promising, and validate what improves production.
Featured by the AI community
How it works
Idea to Production, Systematically
Experiment with confidence.
Integrate discovery, building, and validation.
Discover
Generate environments from relevant ideas.
Build
Turn ideas into a testable changes.
Validate
Stop shipping based on benchmarks and test against actual use.
How Remyx Helps
Learn what works.
Remyx helps engineers test more ideas and helps leads know which ones drive improvements.
For ML Engineers
Keeping up with AI advances is a full-time job. Most publications don't ship production-ready code. Adapting an idea to your codebase takes days before you know if it's worth it.
Searching through ideas → Test relevant hypotheses
Stuck debugging → Measuring results
One idea at a time → Test more and iterate faster
For Team Leads
Your team is testing ideas, but without a way to validate ideas scientifically you can't tell what's working, what's been tried, or what to prioritize next.
Ad-hoc experiments → Structured and repeatable experiments
Overreliance on intuition → Decisions backed by experiment data
One-off tests→ Compounding, reproducible improvements with science
From the Founders
Built by practitioners, for teams building at the frontier
A team of mathematicians and award-winning ML innovators with a decade of experience applying AI in robotics, healthcare, content recommendation, and enterprise data/ml infrastructure.
Salma Mayorquin
CEO & Co-Founder
Applied Mathematics, UC Berkeley. Former Solutions Architect at Databricks advising MLOps strategy from startups to Fortune 500. Award-winning ML innovator recognized by NVIDIA's developer community.
LinkedIn →
Terry Rodriguez
CTO & Co-Founder
UC Berkeley. 10+ years applying ML in healthcare, robotics, and content recommendation at Riot Games, Tubi, Robust.AI. Open-source tools cited by Google DeepMind and used in peer-reviewed research.
LinkedIn →Talks, Pods & Writing
Conference talks, podcast conversations, and field notes on how AI teams go from experiment to production.
🎙Podcast · Heavybit
ExperimentOps: Why MLOps Alone Isn't Enough
Salma unpacks the shift from MLOps to ExperimentOps with stories from Netflix, Yelp, and Stripe on bringing foundation models into production.
Salma MayorquinListen →
🎤Talk · MLOps Community
Beyond Benchmarks: Measuring Success for AI Initiatives
Why offline metrics don't predict online success and how to design evaluation systems that align model development with business outcomes.
Salma MayorquinWatch →
⚡Talk · AG2 Community
GitRank: Automating Paper-to-PR with Agentic Code Execution
Technical deep-dive on how GitRank uses AG2 agents to discover, build, and test research implementations from arXiv to live experiments.
Terry RodriguezWatch →
✍️Blog · Substack
Vibes Don't Scale
Why AI-assisted coding falls apart past the demo stage and what disciplined experimentation looks like when complex code changes are under-determined by simple prompts.
Terry RodriguezRead →
✍️Blog · Substack
Why We Need ExperimentOps
The case for a new operational layer and the announcement of Experiment 2025, the unconference to put it into practice.
Salma MayorquinRead →
✍️Blog · Substack
Building a Knowledge Flywheel
How Terry's research radar caught a new benchmark 4 hours after it hit arXiv and fed it directly into an experiment queue.
Terry RodriguezRead →
📰Feature · Cerebral Valley
Remyx: Your AI Production Assistant
In-depth interview on the mission behind Remyx, how the founders met at Berkeley, and their vision for the next generation of AI engineering tools.
Salma MayorquinRead →
🔌Integration · AG2 Docs
RemyxCodeExecutor: Agentic Code Exploration
Official AG2 integration for discovering papers and executing research code in pre-built environments through an agentic interface.
Terry RodriguezView docs →
🏆Feature · NVIDIA Blog
Smells Like ML: Award-Winning Edge AI Projects
NVIDIA spotlights the founders' earlier work: AI-powered smart mirrors, human movement trackers, and computer vision systems.
Salma & TerryRead →
🎙Podcast · Heavybit
ExperimentOps: Why MLOps Alone Isn't Enough
Salma unpacks the shift from MLOps to ExperimentOps with stories from Netflix, Yelp, and Stripe on bringing foundation models into production.
Salma MayorquinListen →
🎤Talk · MLOps Community
Beyond Benchmarks: Measuring Success for AI Initiatives
Why offline metrics don't predict online success and how to design evaluation systems that align model development with business outcomes.
Salma MayorquinWatch →
⚡Talk · AG2 Community
GitRank: Automating Paper-to-PR with Agentic Code Execution
Technical deep-dive on how GitRank uses AG2 agents to discover, build, and test research implementations from arXiv to live experiments.
Terry RodriguezWatch →
✍️Blog · Substack
Vibes Don't Scale
Why AI-assisted coding falls apart past the demo stage and what disciplined experimentation looks like when complex code changes are under-determined by simple prompts.
Terry RodriguezRead →
✍️Blog · Substack
Why We Need ExperimentOps
The case for a new operational layer and the announcement of Experiment 2025, the unconference to put it into practice.
Salma MayorquinRead →
✍️Blog · Substack
Building a Knowledge Flywheel
How Terry's research radar caught a new benchmark 4 hours after it hit arXiv and fed it directly into an experiment queue.
Terry RodriguezRead →
📰Feature · Cerebral Valley
Remyx: Your AI Production Assistant
In-depth interview on the mission behind Remyx, how the founders met at Berkeley, and their vision for the next generation of AI engineering tools.
Salma MayorquinRead →
🔌Integration · AG2 Docs
RemyxCodeExecutor: Agentic Code Exploration
Official AG2 integration for discovering papers and executing research code in pre-built environments through an agentic interface.
Terry RodriguezView docs →
🏆Feature · NVIDIA Blog
Smells Like ML: Award-Winning Edge AI Projects
NVIDIA spotlights the founders' earlier work: AI-powered smart mirrors, human movement trackers, and computer vision systems.
Salma & TerryRead →
Active in the Research Community Open Source
We contribute open-source tools, datasets, and benchmarks across AI domains and the research community builds on them.
🌐
From landmark paper to open infrastructure
We reproduced DeepMind's SpatialVLM data pipeline entirely with open-source tools, then trained models a fraction of the size, proving small, open models could learn spatial reasoning too. This enabled follow-on NeurIPS publications, new benchmarks, and a wave of open models across the field.
Google DeepMind · CVPR 2024 →
🎓
Datasets adopted by NeurIPS researchers
Oxford and UC Santa Cruz researchers trained and evaluated their models using Remyx open-source datasets.
Oxford / UCSC · NeurIPS 2025 →
📊
Open benchmarks used across research labs
Multiple research teams use Remyx-generated datasets as training data and evaluation benchmarks for advancing AI reasoning capabilities.
arXiv 2025 →
🛠️
VQASynth: Open-source data synthesis pipeline
A toolkit for composing multimodal training datasets, used and extended by researchers worldwide to advance model capabilities.
GitHub · remyxai/VQASynth →
🤝
Partner integration with AG2 (formerly AutoGen)
Co-developed the RemyxCodeExecutor for AG2's agentic framework, enabling researchers to explore and test code from any paper.
AG2 Official Docs →
🏅
NVIDIA hackathon winners & featured makers
Won recognition in NVIDIA's AI at the Edge competition and featured on the NVIDIA blog for open-source edge AI tools built on Jetson.
NVIDIA Developer Blog →
📈
30K+ downloads on community model releases
Lessons learned from publishing open models on Hugging Face and what makes research artifacts actually get adopted by practitioners.
Substack · Myx'd Results →
🌐
From landmark paper to open infrastructure
We reproduced DeepMind's SpatialVLM data pipeline entirely with open-source tools, then trained models a fraction of the size, proving small, open models could learn spatial reasoning too. This enabled follow-on NeurIPS publications, new benchmarks, and a wave of open models across the field.
Google DeepMind · CVPR 2024 →
🎓
Datasets adopted by NeurIPS researchers
Oxford and UC Santa Cruz researchers trained and evaluated their models using Remyx open-source datasets.
Oxford / UCSC · NeurIPS 2025 →
📊
Open benchmarks used across research labs
Multiple research teams use Remyx-generated datasets as training data and evaluation benchmarks for advancing AI reasoning capabilities.
arXiv 2025 →
🛠️
VQASynth: Open-source data synthesis pipeline
A toolkit for composing multimodal training datasets, used and extended by researchers worldwide to advance model capabilities.
GitHub · remyxai/VQASynth →
🤝
Partner integration with AG2 (formerly AutoGen)
Co-developed the RemyxCodeExecutor for AG2's agentic framework, enabling researchers to explore and test code from any paper.
AG2 Official Docs →
🏅
NVIDIA hackathon winners & featured makers
Won recognition in NVIDIA's AI at the Edge competition and featured on the NVIDIA blog for open-source edge AI tools built on Jetson.
NVIDIA Developer Blog →
📈
30K+ downloads on community model releases
Lessons learned from publishing open models on Hugging Face and what makes research artifacts actually get adopted by practitioners.
Substack · Myx'd Results →
Follow us on Substack
Technical deep-dives, experiment logs, and lessons learned from the founders of Remyx AI.
Ready to stop guessing?
Start exploring relevant research and run your first experiment in minutes.