Skip to main content
Yggdrasil supports two ways to define compliance policies: uploading custom regulatory PDFs or selecting from prebuilt frameworks.

​
Prebuilt Frameworks

Choose from three compliance frameworks with ready-to-use rules:

​
AML / FinCEN

11 pre-configured rules covering:
  • Currency Transaction Reports (CTR) thresholds
  • Structuring detection (splitting transactions to avoid reporting)
  • Velocity limits (transaction frequency patterns)
  • Dormant account reactivation
  • Round amount patterns (suspicious even numbers)
  • Balance mismatches
  • Suspicious activity thresholds

​
GDPR

14+ category rules covering:
  • Consent management
  • Data Protection Officer (DPO) requirements
  • Encryption at rest
  • Marketing consent
  • Personal data handling
  • Privacy Impact Assessments (PIAs)
  • Processing records
  • Right of access, erasure, and information
  • Third-country transfer safeguards

​
SOC 2

5 trust principle rules:
  • Security (logical access controls)
  • Availability
  • Confidentiality (encryption requirements)
  • Processing Integrity
  • Privacy
1

Select a prebuilt framework

Choose AML, GDPR, or SOC2 from the policy selection screen. All rules load instantly without AI processing.
2

Review the rule set

Preview the rules before proceeding. You can toggle individual rules on or off later.
3

Continue to data upload

Once selected, proceed to upload your dataset.

​
Custom PDF Upload

Upload any regulatory document to extract custom compliance rules using AI.
1

Upload your PDF

Drag and drop a regulatory PDF (max 500,000 characters). The system supports most PDF formats except scanned images without OCR.
2

AI extraction process

Gemini 2.5 Flash analyzes the document and extracts enforceable rules using the Signal Specificity Framework.Each rule must combine multiple signals to reach a minimum specificity threshold of 2.0:
  • Weak Signals (0.5): Single thresholds, state checks
  • Medium Signals (1.0): Temporal windows, behavioral shifts
  • Strong Signals (2.0+): Multi-condition logic, cross-field dependencies
This prevents broad, single-threshold rules that generate false positives.
3

Review extracted rules

After extraction completes, you’ll see:
  • Total rules extracted
  • Each rule’s severity (CRITICAL, HIGH, MEDIUM)
  • Policy excerpts justifying each rule
  • Compound conditions (AND/OR logic trees)
Transparent design: Extracted rules show the exact policy excerpt they’re derived from, making AI reasoning auditable.

​
What Happens Behind the Scenes

When you upload a PDF:
  1. Text extraction: The system parses your PDF using unpdf (supports up to 500K characters)
  2. AI analysis: Gemini applies the Signal Specificity Framework to identify enforceable rules
  3. Rule structuring: Each rule is formatted as a compound boolean expression (AND/OR trees)
  4. Database storage: Rules are saved to your policy with metadata (severity, thresholds, policy sections)
PDFs must contain extractable text. Scanned images without OCR will fail. If extraction fails, you’ll receive an error message prompting you to check the file format.

​
Rule Quality Standards

All extracted rules (prebuilt or custom) must meet these criteria:
  • Combined specificity ≥ 2.0: No single-condition rules allowed
  • Explainable: Every rule links back to a policy excerpt
  • Executable: Conditions use supported operators (equals, greater_than, contains, IN, BETWEEN, etc.)
  • Testable: Rules can be toggled on/off for testing

​
Supported PDF Formats

✅ Native PDFs with selectable text
✅ Text-based regulatory documents
✅ Policy documents up to 500K characters
❌ Scanned images without OCR
❌ Encrypted or password-protected PDFs

​
Next Steps

After uploading your policy:
  1. Upload your dataset → Data Upload
  2. Review and toggle rules as needed
  3. Proceed to column mapping

Build docs developers (and LLMs) love

Get started for freeTalk to us