cortex-ai-function-studio-skill

Create, evaluate, and optimize custom AI functions using Snowflake Cortex AI Complete. Supports text, image, and document inputs. Use when: building LLM-powered functions, evaluating AI function performance, tuning prompts, selecting models, checking async job status. Triggers: ai function builder, custom ai function, user defined ai function, build my own llm function, evaluate ai function, tune ai function, optimize ai function, demo ai function, resume ai function job, image classification, document analysis, multimodal ai function.

Create a new custom AI function. Supports table-based or manual input specification, single or variant outputs. Direct AI_COMPLETE calls or additional pre- and post-processing.

Quick Start demo: Build a toxicity classifier and evaluate it — the fastest way to experience the core create → evaluate workflow.

Interactive demo: Generate pseudo-labels from a strong teacher model, build a cheap student function, and evaluate accuracy. Showcases pseudo-labeling and teacher-student distillation.

Interactive demo: Build a legal contract field extractor and create a weighted composite metric that scores 4 fields independently. Showcases custom evaluation metrics for multi-field AI functions.

Interactive demo: Extract structured fields from SEC 10-K filing PDFs using multimodal AI, create a custom composite metric for per-field scoring, and evaluate extraction accuracy with per-field analysis.

Interactive demo: Build a policy-conditioned ticket router where a seed prompt performs poorly, then watch prompt optimization close the accuracy gap through prompt evolution and Pareto cost/quality analysis. The canonical demo for prompt optimization.

Interactive demo [Experimental]: Build a PII redaction function using Agent Research mode — the agent searches the web for techniques, proposes SQL UDF architectures with pre/post-processing, and you pick the approach. Then optimize the full function body.

Interactive demos for custom AI functions. Use when: demo, example, walkthrough, show me, how does this work, try it, hands-on, tutorial.

Evaluate an AI function's performance against a labeled dataset using a Python stored procedure.

Optimize an AI function through automated function body optimization, including prompts, model references, and SQL pre/post-processing.

Generate synthetic data or pseudo-label input-only tables for AI function evaluation and optimization. Triggers: generate data, synthetic data, test data, create test cases, make training data, pseudo label, label my input-only table.

Interactive demo: Build a multimodal clothing condition classifier using expert-labeled garment images, then evaluate and optimize across models via GEPA optimization.