Setup Serverless Cloud GPU at runpod.io

## Task Summary

1. Setup serverless GPU at [runpod.io](http://runpod.io/), with the relevant OCR libraries installed
2. Develop a program/script that exposes the OCR services as REST API and host it at the same serverless GPU above

## Problems

Currently we are using OCRmyPDF and EasyOCR library to perform OCR on PDF documents. These OCR libraries are currently running in the same server as the web application. Due to the web application server has no GPU, the OCR processing is slow.

## Objective

The proposed solution is to host these OCR libraries at [runpod.io](http://runpod.io/) using their serverless GPU (https://www.runpod.io/serverless-gpu ) service. In order for the web application server to use the OCR services, a custom OCR REST API need to be developed and exposed at the serveless GPU.

## Libraries Needed To Be Installed in the Serverless GPU

- It is recommended to use Ubuntu 22.04 as the base OS
- OCRmyPDF https://ocrmypdf.readthedocs.io/en/latest/
- EasyOCR https://github.com/JaidedAI/EasyOCR
- Other dependencies libraries
    - Tesseract https://github.com/tesseract-ocr/tesseract
    - Ghostscript (gs) https://www.ghostscript.com/
    - Ps2Pdf https://web.mit.edu/ghostscript/www/Ps2pdf.htm

## OCR REST API Specifications

The server must expose an OCR REST API with the following requests parameters:

- `file_url`, string. URL to the PDF file
- `doc_type`, string. Document Type. Valid values are:
    - BOOKING_FORM Booking Form
    - IDENTITY_DOC Identity Doc
    - LOAN_OFFER_LETTER Loan Offer Letter
- `api_key`, string. API key

The API must then return the following response:

- `status`, string. Status, valid values are: success or fail
- `error`, string. Error, if fail, the error message. If success, this field does not exists
- `output`, string. Output, if success, the OCR text output

### Processing Logic for Different Document Type

1. Booking Form
    1. Use `ps2pdf` to minimize the file
    2. Use `ocrmypdf` to read the minimized file
2. Identity Doc
    1. Use `gs` to convert each page to PNG file
    2. Use `easyocr` to read each PNG file
3. Loan Offer Letter
    1. Use `ocrmypdf` to deskew the file
    2. Use `ocrmypdf` to read the deskewed file

### Remarks

1. Details of the above processing logic is already written in Laravel PHP which you can refer as reference. Request for the source code if you need.
2. The OCR REST API can be written in any language, but Python is preferred. Alternatively if you think hosting the Laravel PHP in the serverless GPU is appropriate (so that we don’t have to rewrite the processing logic again in another language), you can suggest this too.