Architecture

Last verified: 2026-07-26 against dev at f0388abb.

SDP is a shared smart document processing application with a frontend, backend, extractor workers, storage, messaging, and evaluation tooling.

Components

Component	Responsibility
Frontend	React/Vite UI for assessments, validation, extraction/redaction workflows, and organization-scoped API calls.
Backend	FastAPI API for assessments, documents, batches, processors, corrections, exports, tenancy, and auth.
Extractor apps	Azure Function workers that process Service Bus messages and extract structured data.
Service Bus	Queue-based routing from backend processing requests to extractor workers.
Blob storage	Document content and intermediate artifacts.
Database	Assessments, documents, batches, schemas, processors, extracted items, corrections, and status.
Phoenix/Arize	Tracing, dataset, experiment, and evaluator workflows.
Infrastructure repo	Terraform-managed deployed Azure resources.

Runtime Flow

1. A user creates an assessment and uploads documents.
2. Backend stores metadata and document content.
3. A batch/batch run selects documents and a processor.
4. Backend sends work to the extractor queue.
5. Extractor workers process documents and return structured outputs.
6. Backend records extracted items, fields, status, and corrections.
7. Frontend displays review/validation/export workflows.

Design Notes

• Assessments are the top-level organization-scoped unit.
• Processors are schema/extractor combinations.
• Backend registry state and deployed function-app infrastructure are separate; both must be updated for extractor changes.
• BYOK for consumer apps uses server-side forwarding only; browsers must never receive long-lived LLM keys.