Metaphi AI

Environments and evaluations for enterprise work.

Benchmarks

• EnterpriseSWE: long-horizon software engineering on real enterprise systems, verified by execution.
• LH-Bench: long-horizon agents scored on subjective enterprise work by expert rubrics, curated artifacts and human preference. Paper.

Datasets

• frontend-figma-to-code: real Figma designs, built into React and Tailwind by coding agents.
• figma2code-expert-preferences: pairwise expert preferences over those builds.
• remotion-video-gen: courses from source documents, rendered as programmatic video.
• remotion-video-preferences: pairwise expert preferences over those chapters.
• credit-underwriting-preview: business bank statements as PDFs, every transaction labelled by risk analysts.
• credit-underwriting-commercial: the full set, on request.

Environments

Mainframe, Go, Figma, financial document reasoning and content generation, off the shelf at metaphi.ai/environments. Custom environments for partners: agent@metaphi.ai.