Skill UI

Toolkit for Python-based survival and time-to-event modeling with scikit-survival, supporting Cox models, ensemble learners, SVMs, and evaluation metrics for censored data.

Guides agentic engineering efforts where AI agents execute implementation while humans focus on risk-aware evaluation, breaking work into verifiable units, routing models by complexity, and tracking costs and regression checks.

Adds and manages structured evaluation data in Hugging Face model cards, extracting README tables, importing Artificial Analysis benchmarks, and running vLLM/lighteval evaluations to emit model-index metadata.

Run inspect-ai or lighteval evaluations on Hugging Face Hub models locally, guiding backend choice, smoke tests, and GPU fallbacks before handing off remote tasks.

This skill manages the schema for human feedback (annotation configs) within the Arize platform. It allows users to define label types (categorical, continuous, freeform) and programmatically apply these annotations to project spans using the Python SDK. Essential for establishing standardized human review processes and improving model evaluation.

This skill provides comprehensive CLI commands to manage AI integrations within Arize. It covers listing, creating, and retrieving configurations for various major LLM providers (including OpenAI, Anthropic, Gemini, AWS Bedrock, etc.), ensuring secure credential storage and proper setup for model evaluation and usage.

This skill provides comprehensive command-line interface (CLI) functionality for managing the entire lifecycle of experiments within the Arize platform. Users can list, retrieve metadata, create new experiments from various data files (CSV, JSON, Parquet), and export detailed run results for deep analysis. It supports linking experiments to specific dataset versions, making it essential for structured ML model evaluation and performance comparison workflows.

This skill provides comprehensive command-line capabilities for interacting with Arize datasets. Use it to perform dataset CRUD operations (Create, Read, Update), list available datasets, retrieve detailed metadata, and export large volumes of example data. It is essential for managing the lifecycle of versioned datasets used in model evaluation and machine learning experimentation. Supports file formats like CSV, JSON, and Parquet.

Defines and runs LLM-as-judge evaluators on Arize via ax commands, handling evaluator creation, continuous monitoring, column mapping, trigger runs, and evaluation spans or experiments with configurable templates, models, and query filters.