Expert support for customer training needs

micro1 is a data and AI research company that helps AI labs and enterprises train and evaluate models and agents. It combines domain experts with a data platform to produce training data, build reinforcement learning environments, and evaluate AI performance on real-world tasks.

Under the proposed model, Fireworks would introduce customers to micro1 when a training project needs additional data or human expertise. micro1 would work directly with the customer to source or create training data, label existing examples, and review model outputs. For reinforcement learning projects, the scope could also include custom environments: tasks and feedback that help a model learn the customer’s workflow.

This would extend the support available to Fireworks customers as they adapt models to their businesses. micro1 would own the agreed data and expert work, while Fireworks would provide the training and inference platform. Customers would have a clearer path from identifying a model’s weaknesses to improving it and putting it into production.

How an engagement would work

Scope the customer’s task. Fireworks, micro1, and the customer would identify the target workflow, available data, and model behaviors that need improvement. Together, they would agree on the training approach, the expert work required, and how results will be assessed.

Build the data and feedback. micro1 would work directly with the customer to deliver the agreed training inputs. This could mean purchasing suitable data, commissioning new examples, or having experts label and assess the customer’s own data. For an RL project, the teams would also define the tasks and scoring criteria needed for the training workflow.

Train, evaluate, and deploy on Fireworks. The customer would use the resulting data or feedback with a supported Fireworks training method, evaluate the model against the agreed goals, and deploy it on Fireworks. The engagement scope would define responsibility for configuring training runs and evaluations. Results could then inform the next round of expert work.

Bring your use case

If you’re developing a model for a specific task, connect with the Fireworks and micro1 teams to discuss your training needs. The starting point is the customer’s goal: what the model should do, where it falls short today, and what data or expertise would help close that gap.