Documents go through a defined pipeline with multiple stages at amaise:
CREATED → OCR → SEGMENTATION → SEGMENTATION REVIEW → SPLITTING → INDEXING → EXTRACTION → ANALYSIS → ANSWERING → CASE REVIEW → READY
Files that are not PDFs (for example Word documents or images) are first converted to PDF, if these file types are enabled for your workspace. Archives, folders and emails are unpacked, and each file inside them is added separately. The file you uploaded is kept unchanged.
Key features:
Idempotent workers: Each stage is handled by a standalone, stateless worker. Processing can be safely repeated in case of errors.
Asynchronous communication: Workers communicate via message queues (SQS). Each worker processes one task at a time.
Tenant separation: Each task is assigned to a specific tenant. The same tenant isolation controls apply as in the rest of the application.
Encrypted storage: Documents are stored in S3 with tenant-specific encryption keys.
Human review stages: Segmentation review and case review are built-in checkpoints where users verify and correct AI results; when a review is required is configured per tenant.