CoreWeave launched ARIA, an AI research agent inside Weights & Biases. It analyzes experiments, builds live dashboards, and launches the next run automatically. ARIA forms hypotheses, writes configs, and runs experiments via W&B Launch on team compute. It evaluates results, updates workspaces with comparison panels, and drafts reports. ARIA creates shareable workspaces, panels, and reports to back up findings. It selects the right panel type for the question and filters to the most important runs. ARIA loads full project context, including code, logs, metrics, and artifacts. It can span projects and reference teammates’ experiments. ARIA is trained on Weights & Biases platform nuances for efficient querying. It is available now.
What changed
Researchers manually analyzed experiments, built dashboards, and launched runs in Weights & Biases.
What it unlocks
Automating the full research loop from hypothesis to next experiment launch within Weights & Biases.
What you need to act on it
- a Weights & Biases workspace
- ARIA access enabled
Sources