Name: LLMWhisperer – OCR & Document Parsing | Free · $5/mo | API.market
Brand: unstract
SKU: api-unstract-extract
Availability: InStock
Author: API Market

Shashank Agarwal

# LLMWhisperer – Document Parsing & OCR API

This API extracts text and (optionally) form element data from a wide range of documents—PDFs, PDF forms, images (JPEG, PNG, TIFF), MS Office files (Word, Excel, PowerPoint), LibreOffice files, and more. It’s designed for AI/LLM workflows (e.g., ChatGPT, embeddings) and supports scenarios like **resume parsing**, **document parsing**, **PDF-to-text for LLMs**, and **PPT-to-text extraction**.  

Once a document is submitted (via URL or file upload), you get back a job ID (`whisper_hash`). You can retrieve the parsed text (and metadata if needed) by calling the retrieval endpoint.  

---

## Key Features

1. **OCR for Scanned Documents and Images**  
   - Parses scans, photos, or other non-text PDFs into machine-readable text.  
   - Supports handwriting recognition (in higher modes).

2. **Form Mode with Checkbox/Radio Detection**  
   - Ideal for PDF forms, questionnaires, or surveys.  
   - Detects if checkboxes or radio buttons are checked.

3. **Layout-Preserving Extraction**  
   - Optionally preserve layout (line breaks, spacing) to maintain textual structure for better LLM performance.

4. **Multi-Language Support**  
   - Low Cost mode supports 120+ languages (English by default).  
   - High Quality modes support 300+ languages.

5. **Page-Level Control & Separation**  
   - Define page ranges and custom page separators.  
   - Retrieve multi-page documents in a consistent format for further AI analysis.

6. **Image Preprocessing**  
   - Median filter and Gaussian blur (in Low Cost mode) to enhance OCR results on noisy scans.

7. **Privacy by Design**  
   - Extraction results can be retrieved only once; then they’re deleted permanently.  
   - Retrieval calls cost 0 units (free).

---

## Common Use Cases

1. **Resume Parsing & HR Automation**  
   - Convert resumes in PDF, Word, or image format to raw text for AI-based entity extraction or classification.  
   - Minimize manual data entry and speed up hiring processes.

2. **Document Parsing & Data Extraction**  
   - Extract contract text, legal documents, or corporate memos into structured or plain text.  
   - Feed results into Large Language Models for summarization or Q&A.

3. **PPT/PowerPoint to Text for LLM**  
   - Convert PowerPoint slides into text for AI summarization or indexing.  
   - Maintain layout cues to preserve bullet points and structure for NLP tasks.

4. **PDF-to-Text for LLM**  
   - Parse scanned or native PDFs and produce LLM-ready text.  
   - Optionally preserve the page structure and layout for maximum LLM accuracy.

5. **Form Field Recognition**  
   - Detect checkboxes, radio buttons, and other form elements in questionnaires, surveys, or application forms.  
   - Useful for automating data entry and analysis.

6. **Multi-Language Document Processing**  
   - Process scanned documents in various languages and scripts.  
   - Ideal for global organizations handling diverse language documents.

---

## Modes & Pricing

| Mode          | OCR Method                           | Price/Call | Recommended For                                                       |
|---------------|---------------------------------------|------------|-----------------------------------------------------------------------|
| **low_cost**  | Basic OCR + optional image preprocessing  | 1 API unit     | PDFs (native or scanned), images with good quality, cost-sensitive apps |
| **high_quality** | Advanced OCR with AI-based enhancement | 2 API units     | Lower-quality scans, handwritten documents, complex backgrounds       |
| **form**      | High Quality + Form Field Detection   | 3 API units     | PDF forms, checkboxes, radio buttons, surveys, complex handwriting    |

### Retrieval Calls
- **0 units** (free).  
- Processed documents are deleted after a single retrieval for privacy/security.

---

## Detailed Mode Comparison

| Mode                                          | Low Cost (low_cost)| High Quality (high_quality) | High Quality + Forms (form) |
|--------------------------------------------------|-----------|-------------|----------------------|
| **Checkbox / Radio button detection**            | No        | No          | Yes                  |
| **Line reproduction in output**                  | Yes       | Yes         | Yes                  |
| **Supported Languages**                          | 120+      | 300+        | 300+                 |
| **Image preprocessing (median filter, blur)**    | Yes       | No          | No                   |
| **Layout preserving output**                     | Yes       | Yes         | Yes                  |
| **Extraction performance**                       | Fast      | Medium      | Medium               |
| **Handwriting recognition**                      | Basic     | Advanced    | Advanced             |
| **AI/ML-based enhancement**                      | No        | Yes         | Yes                  |
| **Rotation/skew compensation**                   | No        | Yes         | Yes                  |
| **Auto-repair PDFs**                             | Yes       | Yes         | Yes                  |
| **Dense text content**                           | Very Good | Very Good   | Very Good            |
| **High entropy content**                         | Very Good | Very Good   | Very Good            |

---

## Quick Start

1. **Obtain API Key**  
   - Use `x-magicapi-key: API_MARKET_API_KEY` in your requests.

2. **Submit Document**  
   - **Upload** the file in binary form (`application/octet-stream`).  
   - **OR** pass a public URL (`text/plain`).
   - If submitting a document via URL, set url_in_post=true to indicate URL submission.

3. **Receive `whisper_hash`**  
   - The API responds with `whisper_hash` indicating the job is accepted and being processed asynchronously.

4. **Retrieve Result**  
   - Use `GET /whisper-retrieve` with the same `whisper_hash` to get the extracted text.  
   - If `text_only=true`, you'll receive just the plain text. Otherwise, you get more metadata (e.g., form fields).

---

## Example Requests

### 1. Submit via URL

```bash
curl -X 'POST' \
  'https://api.magicapi.dev/api/v1/unstract/extract/whisper?mode=low_cost&output_mode=layout_preserving&page_seperator=%3C%3C%3C&line_splitter_tolerance=0.4&line_splitter_strategy=left-priority&horizontal_stretch_factor=1&mark_vertical_lines=false&mark_horizontal_lines=false&lang=eng&url_in_post=true' \
  -H 'accept: application/json' \
  -H 'x-magicapi-key: API_MARKET_API_KEY' \
  -H 'Content-Type: text/plain' \
  -d 'https://www.sbs.ox.ac.uk/sites/default/files/2019-01/cv-template.pdf'
```

**Response**:
```json
{
  "message": "Whisper Job Accepted",
  "status": "processing",
  "whisper_hash": "4cd4df6e|42ed92497e8751b88b3fd00789421b4f"
}
```

### 2. Submit via File Upload

```bash
curl -X 'POST' \
  'https://api.magicapi.dev/api/v1/unstract/extract/whisper?mode=low_cost&output_mode=layout_preserving&page_seperator=%3C%3C%3C&line_splitter_tolerance=0.4&line_splitter_strategy=left-priority&horizontal_stretch_factor=1&mark_vertical_lines=false&mark_horizontal_lines=false&lang=eng' \
  -H 'accept: application/json' \
  -H 'x-magicapi-key: API_MARKET_API_KEY' \
  -H 'Content-Type: application/octet-stream' \
  --data-binary '/path_to_file.pdf'
```

**Response**:
```json
{
  "message": "Whisper Job Accepted",
  "status": "processing",
  "whisper_hash": "4cd4df6e|42ed92497e8751b88b3fd00789421b4f"
}
```

---

## Retrieving the Parsed Document

```bash
curl -X 'GET' \
  'https://api.magicapi.dev/api/v1/unstract/extract/whisper-retrieve?whisper_hash=4cd4df6e%7C42ed92497e8751b88b3fd00789421b4f&text_only=true' \
  -H 'accept: application/json' \
  -H 'x-magicapi-key: API_MARKET_API_KEY'
```

- **`text_only=true`**: Returns only plain text.  
- **`text_only=false`**: Returns extra metadata (e.g., checkbox states, bounding boxes).  

**Response (text only)**:
```json
{
  "text": "Extracted text from your document ..."
}
```

**Response (with metadata)**:
```json
{
  "whisper_hash": "4cd4df6e|42ed92497e8751b88b3fd00789421b4f",
  "data": {
    "text": "Extracted text ...",
    "metadata": {
      "forms": [
         { "field_name": "consent", "is_checked": true },
         ...
      ]
    }
  }
}
```

> **Note**: Once retrieved, the document is removed from our servers.

---

## Pricing Summary

- **Low Cost Mode**: 1 unit/call  
  - Good for **simple PDFs**, **cost-sensitive** workflows, and **native text** PDFs.  
  - Basic handwriting support.

- **High Quality Mode**: 2 units/call  
  - Handles **challenging scans**, **low-quality** images, and **handwritten text** more accurately.

- **Form Mode**: 3 units/call  
  - Includes form element detection (checkboxes, radio buttons) and advanced OCR.  
  - Great for **PDF forms** (applications, surveys, medical forms).

**Retrieval Calls**: Free (0 units)

---

## Best Practices

- **Choose the Right Mode**  
  - Use `low_cost` when documents are clean, typical PDFs or you’re budget-conscious.  
  - Use `high_quality` for more complex, lower-quality, or partially handwritten documents.  
  - Use `form` for documents with checkboxes, radio buttons, or advanced form fields.

- **Preserve Layout for LLMs**  
  - Set `output_mode=layout_preserving` to help LLMs understand textual structure better, crucial for extracting tables or bulleted lists.

- **Reduce Token Usage**  
  - Consider trimming or summarizing the text before sending it to an LLM to reduce token consumption.

- **Check Language**  
  - Set `lang` parameter if you know the document language to improve OCR accuracy.

---

## Conclusion

LLMWhisperer simplifies **document parsing**, **resume parsing**, **OCR extraction**, and **PPT/Excel/Word-to-text** conversion for advanced AI workflows. By capturing layout, handling forms, and providing robust multi-language OCR, it powers better results from Large Language Models.  

Use it to parse and analyze PDFs, images, Office files, or LibreOffice documents with minimal overhead. Submit your document, retrieve the processed text once, and feed it directly into your LLM or AI pipeline.  

```Note: When submitting a document via URL, always include url_in_post=true to indicate the API should parse the document from the provided URL.```

Extract text & form data from PDFs, resumes, Office files, and images for your LLM workflows. Free 100 calls/mo, paid from $5/mo. Start free on API.market.

LLMWhisperer – OCR & Document Parsing | Free · $5/mo | API.market

FREE

Unstract

BUILD

GROW

ULTRA

MEGA

Efficiently convert multi-page PDFs, form-based documents, and PowerPoint slides into high-fidelity text that’s ready to power your AI or LLM solutions.