Multi-language extraction
Arabic (native) and English out of the box, with configurable extractors for other RTL and LTR languages. Print, typed, and handwriting supported.
- Arabic
- English
- Handwriting
Data & AI · Document AI · Smart OCR
Documents in. Structured records out.
Arabic and English extraction from IDs, invoices, contracts and handwriting: structured JSON in S3, feeding Bedrock knowledge bases.
Smart OCR is Phoenix's productized document-AI pipeline, live on AWS Marketplace. It reads Arabic and English documents, including handwriting, and delivers structured JSON in Amazon S3 that downstream systems, RAG knowledge bases, and dashboards can consume directly.
Deployed to customer AWS accounts, so the documents never leave the customer's data boundary. Sold as a product, tuned per customer on top of the same core pipeline.
What's in scope.
Document types
IDs and passports, invoices and receipts, contracts, KYC packs, medical documents, government forms, tabular data: extractor templates per doc-type.
- IDs
- Invoices
- Contracts
- Forms
Structured JSON to S3
Every extracted document becomes a normalised JSON record in Amazon S3 with confidence scores per field: feed BI, integrate to SAP or CRM, index into Bedrock.
- S3
- JSON
- Confidence scores
Bedrock-ready
Extraction pipeline plugs directly into Amazon Bedrock knowledge bases: the same document corpus becomes a RAG source for enterprise copilots.
- Bedrock KBs
- RAG-ready
The approach.
-
01
Scope
Doc types, languages, volume, target downstream systems: priced against a per-page tier.
-
02
Deploy
Product installed into the customer's own AWS account: data stays inside their boundary.
-
03
Tune
Extractor templates configured per doc type; sample corpus benchmarked to accuracy targets.
-
04
Scale
Volume ramp, integrations wired to SAP / CRM / knowledge base, ongoing accuracy monitoring.
Named deliverables.
Every engagement lands specific artefacts, not slides.
- Smart OCR product deployed inside your AWS account
- Per-document-type extractor templates configured and benchmarked
- Structured JSON output landing in your Amazon S3 buckets
- Downstream integration to SAP, CRM, BI or Bedrock knowledge base
- Accuracy dashboard with per-field confidence scoring
- Ongoing tuning as new document types are onboarded
Frequently asked
How is Smart OCR different from open-source OCR or Amazon Textract?
Smart OCR is a productized pipeline tuned for Arabic, English and handwriting, with per-document-type extractor templates (IDs, invoices, contracts, KYC), confidence scoring, and JSON-in-S3 output ready for downstream systems. It's built on AWS services (including Textract where it fits) but delivered as a product, not a toolkit.
Does the document data leave our AWS account?
No. Smart OCR deploys into the customer's own AWS account. Documents are extracted, JSON output lands in the customer's own S3 buckets, and everything stays inside the customer's data boundary.
Which document types are supported out of the box?
National IDs and passports, invoices and receipts, contracts, KYC packs, medical documents, government forms and tabular data. New document types are added by configuring an extractor template: usually days, not weeks.
How is accuracy monitored and improved?
Every extracted field carries a confidence score. Per-doc-type dashboards report accuracy and low-confidence rates. Human-in-the-loop review can be wired in for high-value docs, and low-confidence samples feed back into extractor tuning.
Can Smart OCR feed a Bedrock RAG copilot?
Yes, that's the intended shape. Extracted JSON is stored in S3 in an already-normalised form, so plugging it into Amazon Bedrock Knowledge Bases is a straight indexing step. Same corpus becomes the ground truth for enterprise copilots.
The earliest conversations are usually the most useful.
Whether you're scoping an SAP move to cloud, restarting a stalled programme, or just trying to figure out where data and AI fit, start with a conversation.
