Document Intelligence

Upload an Indonesian document — KTP, NPWP, NIB, faktur pajak, invoice, contract, akta — get back structured fields with confidence scores and bounding boxes.

Branded as Quantum AI in customer-facing surfaces. The internal model stack is intentionally not exposed.

What it does

Capability	API
Upload + auto-classify	`POST /api/documents/upload` (multipart)
Extract by template	Auto-applied when classification has high confidence
Templates (custom schemas)	`POST /api/templates` — define your own field set
Re-extract with overrides	`POST /api/documents/{id}/reextract`
Analyst correction	`POST /api/documents/{id}/extractions` (manual fields)
Export	`GET /api/documents/{id}/export` (JSON or CSV)
Webhooks	Push on `extracted`, `requires_review`, `failed`

Supported document types

Group	Types
Indonesian IDs	`ktp` · `npwp` · `kartu_keluarga` · `sim` · `passport` · `nib` · `bpkb`
Civil registry	`akta_lahir` · `surat_nikah`
Commercial	`invoice` · `purchase_order` · `delivery_note` · `receipt` · `faktur_pajak` · `slip_gaji` · `contract`
Financial	`rekening_koran` · `financial_statement`
Generic	`generic_ocr` · `other`

Each type has a curated extraction template — known field set, known formats, known validation rules. For documents that don't fit a known template, generic_ocr extracts free-form text + a best-effort key-value pass.

Use the right product for the right document

For KTP identity verification (anti-spoof + tamper + face enrollment), use Identity Platform → KTP capture. Document Intelligence's KTP type returns the OCR fields but does not run liveness or face enrollment. Two products, two purposes.

Core concepts

Concept	What it is
Document	A single uploaded file. Has a type (classified or hint-provided), a status, and ≥0 extractions.
Extraction	A structured field set with confidence + bounding boxes. Multiple per document (model-extracted, template-extracted, analyst-corrected).
Template	A user-defined schema for a custom document type. Specifies fields, types, and required-ness.
Confidence	0–100. Below the org's `requiresReviewThreshold` (default 70), the document is queued for analyst review.

Common integration shape

1
POST/api/documents/upload
Upload — multipart with `file`. Returns `{ document: { id, status: "queued" } }`.
2
Async processing.
Classify → Extract → `status: "extracted"` or `"requires_review"`.
3
Get result.
Subscribe to the `document.extracted` webhook, or poll `GET /api/documents/{id}` until status flips.
4
GET/api/documents/{id}/extractions
Read structured fields — or rely on the webhook payload.

Endpoints at a glance

Group	Endpoints
Documents	`GET/POST /api/documents` · `POST /api/documents/upload` · `POST /api/documents/upload-json` · `GET /api/documents/{id}`
Document actions	`GET /api/documents/{id}/download` · `GET /api/documents/{id}/export` · `POST /api/documents/{id}/erase` · `POST /api/documents/{id}/reextract`
Extractions	`GET/POST /api/documents/{id}/extractions` · `GET /api/extractions/{id}`
Templates	`GET/POST /api/templates` · `GET/PATCH/DELETE /api/templates/{id}`
Search	`GET /api/search?q=...` — min 3 chars; case-insensitive across filename, customer name, extracted-field values
Webhooks	`GET/POST /api/webhooks` · `POST /api/webhooks/{id}/test`
API keys	`GET/POST /api/api-keys`
Health	`GET /api/healthz`
Admin / audit (dashboard)	`GET /api/activity` · `GET /api/audit-log` · `GET /api/metrics` · `GET/POST /api/roles` · `GET/POST /api/members` · `GET/POST /api/invites` · `GET/POST /api/integrations`

Routes marked (dashboard) are intended for analyst UI use and require an internal-scope key.

Production considerations

Concern	Answer
Data residency	Uploaded PDFs and extracted JSON live in `id-jkt-1`. PDFs are sealed at rest with AES-256-GCM in org-isolated storage. Never replicated cross-border.
Retention	Original PDFs: configurable per org (default 90 days). Extracted JSON: retained indefinitely (regulated record). Set retention via `PATCH /api/organization`.
Model / parser	Per-document-type parser pipeline. AI fallback runs as Quantum AI; internal model stack not exposed. Premium tier (>20MB files, faster) on enterprise plans.
Rate limits	Upload: 60/min/org (sandbox), 600/min (production). Re-extraction: 10/min/document. Search: 30/min.
Idempotency	Upload deduplicates on SHA-256 of file content — identical bytes within 24h return the original `documentId` without re-charging. Re-extractions intentionally don't dedup; each call creates a new extraction row (so you can compare prompt versions).
Audit	Every upload, extraction, template change, erase, member action audit-logged. Immutable, 7 years. Filter via `GET /api/audit-log`.
Erasure	`POST /api/documents/{id}/erase` cryptographically shreds the sealed PDF + clears extraction PII (UU PDP-compliant).
Webhook signing	HMAC-SHA256 over the raw body, header `X-DocInt-Signature: sha256=<hex>`. Replay protection: include `X-DocInt-Timestamp` in your signed-payload check + reject deliveries >5min old.