USE CASES

Practical AI classification workflows

Typed outputs work best when the same decision must be made repeatedly. These patterns show how to shape state and questions for common evaluation, moderation, and labeling tasks.

Resume screening

Rank candidates against role-specific levels without parsing free-form essays.

State: Candidate experience, skills, role requirements, and evidence from the resume.
Questions: Use Score for job fit, then Noul for a clear advance-or-review decision.

Support quality assurance

Audit support conversations consistently across a large ticket queue.

State: Transcript, policy requirements, resolution status, and customer sentiment signals.
Questions: Use Noul for policy compliance and Choice for final customer sentiment.

LLM guardrails

Classify risky prompts before they reach a generative model.

State: The user prompt, product policy, trust tier, and relevant conversation context.
Questions: Use Noul for jailbreak detection and Choice for allow, sanitize, block, or escalate.

Content moderation

Apply explicit marketplace or community rules to posts and listings.

State: Content text, metadata, policy definitions, and account history that is safe to use.
Questions: Use Choice for the policy category and Noul for whether enforcement is required.

Survey response coding

Turn open-text responses into stable categories for quantitative analysis.

State: One response, the survey question, and a controlled taxonomy of themes.
Questions: Use Choice for the primary theme and Score for response quality or urgency.

Data labeling

Bootstrap structured labels while keeping uncertain records in a review queue.

State: The source record, label definitions, edge-case policy, and known examples.
Questions: Use any typed primitive, then route low-confidence answers to a human reviewer.

Build a classification workflow