Ctrl AI

Building Units

Six ways to create reasoning units in Ctrl AI.

Units can be created through six different paths, from fully AI-assisted to fully manual.

Path 1: AI-Assisted from Questions (Build Page)

The fastest way to create units. Go to Build, select a domain, and type a question that represents the reasoning you want to capture.

  1. Type a question (e.g., "How do you assess credit risk?")
  2. The AI generates 3-5 candidate units with full formal definitions
  3. Review candidates and approve the useful ones
  4. Approved units enter the verification queue

Path 2: Document Import (Ingest Page)

Extract reasoning from existing documents (policies, procedures, guidelines).

  1. Go to Ingest, select a domain
  2. Paste or upload document text
  3. The system chunks the document by headings
  4. For each chunk, the AI identifies which unit type best captures the knowledge
  5. Review and approve extracted candidates

Path 3: Template Import

Start from pre-built templates for common enterprise processes.

  1. Go to Build > Templates
  2. Browse available templates (Credit Risk, KYC Protocol, Fraud Detection, etc.)
  3. Import a template into your domain
  4. Customize to match your specific requirements

Path 4: Manual Creation

Full control over every aspect of the unit.

  1. Go to Units > New
  2. Fill in: name, type, Given/When/Then, I/O schema
  3. Add formal definition elements (steps, criteria, conditions)
  4. Submit for verification

Path 5: Save from Conversation

Capture useful reasoning that emerges during natural usage.

  1. Ask a question in Ask
  2. Review the AI's response
  3. Hover over a useful segment and click Save as Unit
  4. The AI formalizes the segment into a proper unit
  5. Review and submit for verification

Path 6: Auto-Generate from Coverage Gaps

Let the system identify what's missing and fill the gaps.

  1. Every query that lacks coverage is logged in the demand tracker
  2. View top gaps in the Dashboard > Coverage Gaps section
  3. Click Auto-Generate Units to create candidates for the most-requested missing outputs
  4. Review generated candidates in the verification queue

After Creation

All paths produce units with status: "candidate". These enter the verification queue at /Review and must be verified before they appear in inference results.

See Verification for the verification workflow.

On this page