Skip to main content

​
Architecture

Yggdrasil uses Supabase PostgreSQL with Row-Level Security (RLS) enabled on all tables. Every table is filtered by auth.uid() via RLS policies, ensuring that each user only sees their own data.

​
Database Relationships

auth.users
    └── policies (user_id)
        └── rules (policy_id)
    └── scans (user_id, policy_id)
        ├── violations (scan_id)
        └── pii_findings (scan_id)

​
Core Tables

  • policies — Policy metadata for AML, GDPR, SOC2, or custom PDF policies
  • rules — Extracted compliance rules with compound condition logic and Bayesian feedback counters
  • scans — Scan execution records with compliance scores and trend tracking
  • violations — Detected compliance violations with evidence and explanations
  • pii_findings — PII detection results from uploaded datasets

​
Row-Level Security

All tables use RLS policies that filter by auth.uid(). This means:
  • Users can only access their own policies, scans, and violations
  • No cross-user data leakage is possible at the database level
  • RLS policies are enforced at the PostgreSQL level, not in application code

​
Storage Strategy

In-Memory vs. Persistent StorageCSV data and column mappings are kept in server memory during the audit flow and are not persisted to the database. Only the final scan results (violations, compliance scores, PII findings) are stored in PostgreSQL.

​
Design Rationale

  • JSONB for conditions — Rules support arbitrarily nested AND/OR logic trees. JSONB allows flexible schema without join overhead.
  • Score history on scans — Tracking per-review score changes enables compliance trend visualization without a separate table.
  • Bayesian counters on rules — approved_count and false_positive_count enable the precision formula (1 + TP) / (2 + TP + FP) that improves rule confidence over time.
  • PII findings linked by upload_id — PII detection runs at upload time (before a scan exists), so findings are initially stored by upload_id and linked to scan_id after scan completion.

Build docs developers (and LLMs) love

Get started for freeTalk to us