TracegenceNever fear a document audit again
Create a tenantCustomer login
How it works

Six stages, one audit log.

Between 'a supplier sent us a PDF' and 'the auditor saw the proof' are six deterministic stages. Every stage is observable, retry-safe, and writes a row to the audit log.

Team reviewing dashboards and process flow on a laptop
AUDIT LOG · APPEND-ONLY↓Capture⌕Classify✓Validate⚐Decide◷Score🔒Audit
  1. 01

    Capture

    S3 direct upload from the portal · tokenised no-login URL the supplier opens from email · inbound SES (P1) · CSV / ZIP bulk import (P1).

    POST /documents · POST /public/supplier-upload/<token>
  2. 02

    Classify & extract

    AWS Textract pulls forms + tables. Vertex AI Gemini decides the document type and pulls structured fields. Confidence-band routing: ≥0.85 auto-accept, <0.60 reject, in-between → human queue.

    apps/pipeline/textract.py · apps/pipeline/bedrock.py
  3. 03

    Validate

    Five rule types built in - required_fields, field_pattern, expiry_in_future, signed, rag_check. RAG retrieves top-K rulebook chunks and asks Gemini to cite the failing clause.

    apps/validation/engine.py · apps/rag/validator.py
  4. 04

    Decide

    Sequential or parallel approvals. Every action HMAC-signed and chained so a single tampered row breaks the whole chain at audit time.

    apps/workflows/services.py
  5. 05

    Score

    Tenant compliance score (0–100) recomputed on every status change. History kept as snapshots so you can answer "what was our score last March?".

    apps/scoring/services.py
  6. 06

    Audit

    Append-only audit log with a Postgres trigger that physically blocks UPDATE / DELETE. CSV export, Object Lock retention on Enterprise.

    apps/audit_log/views.py

Try it on your own document.

Spin up a tenant, upload a CoA, watch the engine cite the failing clause.

Create a tenantCustomer login