Forge OCR Server v0.1.0

Multi-layer document OCR with consensus and validation

Active Layers

Document Types

50 MB

Max Upload

~1-3s

Avg Response

API Endpoints

POST /api/v1/ocr/{doc_type} OCR any document (multipart)

POST /api/v1/ocr/{doc_type}/json OCR any document (base64 JSON)

POST /api/v1/pan/ocr PAN card OCR (multipart)

POST /api/v1/pan/ocr/json PAN card OCR (base64 JSON)

GET /api/v1/documents List document types

GET /api/v1/pan/config Default PAN layer config

GET /api/v1/layers Available OCR layers

GET /api/v1/health Health check

Supported Documents

PAN Card (India)

doc_type: pan_card

pan_number PAN Number required
name Name required
fathers_name Father's Name required
date_of_birth Date of Birth required

Aadhaar Card (UIDAI, India)

doc_type: aadhaar_card

aadhaar_number Aadhaar Number required
name Name required
date_of_birth Date of Birth required
gender Gender required
address Address optional
masked_number Masked Aadhaar Number optional

OCR Layers

ID	Description	Weight	PSM	Preprocess	Status
`ocrs_deep_learning`	OCRS Deep Learning OCR (PSM 3 + denoise) — best accuracy	0.90	3	grayscale, denoise	active
`tesseract`	Tesseract LSTM (PSM 3, grayscale) — reliable PAN/DOB	0.70	3	grayscale	active
`onnx_ocr`	ONNX OCR (PSM 6 + binarize) — slow on large images, disabled	0.85	6	grayscale, binarize	disabled
`llm_anthropic`	LLM Vision (PSM 11 sparse) — garbles names, disabled	0.95	11	none	disabled

Quick Start

# Health check curl https://forge-ocr.sachinkumarskrose.workers.dev/api/v1/health # PAN card OCR (multipart) curl -X POST https://forge-ocr.sachinkumarskrose.workers.dev/api/v1/pan/ocr \ -F "image=@pan_card.jpg" # With layer details curl -X POST https://forge-ocr.sachinkumarskrose.workers.dev/api/v1/pan/ocr \ -F "image=@pan_card.jpg" \ -F "include_layers=true" # Single layer override (fastest) curl -X POST https://forge-ocr.sachinkumarskrose.workers.dev/api/v1/pan/ocr \ -F "image=@pan_card.jpg" \ -F 'config=[{"id":"ocrs_deep_learning","description":"OCRS","weight":0.9,"psm":3,"preprocess":{"grayscale":true,"binarize":false,"denoise":true,"deskew":false},"enabled":true}]' # Generic route (any document type) curl -X POST https://forge-ocr.sachinkumarskrose.workers.dev/api/v1/ocr/pan_card \ -F "image=@pan_card.jpg"

How It Works

Upload image (multipart or base64 JSON)
Multiple OCR layers process image in parallel
Fields parsed from raw text using document-specific regex
Results merged via weighted consensus per field
Each field validated (format, checksum, date range)
Response: extracted fields, confidence score, review recommendation