Ocr

/llamafarm:ocr - Optical Character Recognition

Extract text from images, scanned PDFs, and documents using configurable OCR backends.

Usage

/llamafarm:ocr                      # Extract text from a file (prompts for path)
/llamafarm:ocr <file_path>          # Extract text from specified file
/llamafarm:ocr --backend surya      # Use a specific OCR backend
/llamafarm:ocr --lang en,de         # Specify languages for extraction

What This Command Does

1. Locates the file — verifies the file exists and is a supported format
2. Selects a backend — recommends the best OCR engine for the file type
3. Extracts text — runs OCR and returns structured text output
4. Presents results — extracted text with confidence and layout info
5. Suggests next steps — downstream tasks like classification or indexing

Implementation

When the user runs /llamafarm:ocr, follow these steps:

Step 1: Gather Input

Ask the user:

1. What file do you want to extract text from? (file path)
2. What language(s) is the document in? (default: English)
3. Do you have a backend preference? (default: surya)

Step 2: Verify File Exists

ls -la "<file_path>"
file "<file_path>"

Supported formats: .png, .jpg, .jpeg, .tiff, .bmp, .pdf (scanned), .webp

Step 3: Select Backend

Default recommendation: surya for most documents.

Step 4: Execute OCR

curl -X POST http://localhost:14345/v1/ocr \
  -H "Content-Type: application/json" \
  -d '{
    "file_path": "<absolute_file_path>",
    "backend": "surya",
    "languages": ["en"]
  }'

For multi-page PDFs:

curl -X POST http://localhost:14345/v1/ocr \
  -H "Content-Type: application/json" \
  -d '{
    "file_path": "<absolute_file_path>",
    "backend": "surya",
    "languages": ["en"],
    "pages": "all"
  }'

Step 5: Present Results

OCR Extraction Results
======================

File: invoice_scan.pdf
Backend: surya
Pages: 3
Language: en

Extracted Text (Page 1):
-----------------------
ACME Corporation
Invoice #12345
Date: 2024-01-15
...

Confidence: 94.2%
Characters extracted: 2,847
Processing time: 1.3s

Show:

• Extracted text per page
• Overall confidence score
• Character count and processing time
• Any warnings (low confidence regions, skipped pages)

Step 6: Suggest Next Steps

Based on the extracted content, suggest:

Next Steps
==========

1. Classify this document:
   /llamafarm:classify
   Use the extracted text with labels like "invoice", "receipt", "contract"

2. Extract entities (names, dates, amounts):
   Use NER on the extracted text via the ML runtime

3. Index in RAG for search:
   Upload the extracted text to a LlamaFarm dataset for retrieval

4. Run anomaly detection on extracted values:
   /llamafarm:anomaly detect

Error Handling

Skills to Load

• ml-nlp - NLP pipeline reference for post-OCR processing

Related Commands

• /llamafarm:classify - Classify extracted text
• /llamafarm:ml-status - Check ML service health