Achyuth Kumar presents his technical submission for the Book Cover Validation Automation project. This video outlines the key components of the automation process.
Chapters
00:00
Introduction
Greeting and personal introduction
00:13
Project Overview
Transcript
Copy
00:01
Hi this is Achit Kumar.
00:02
This is my submission for round two which is book cover validation automation.
00:07
So I built this automation in N8N for book cover validation.
00:12
So I used 8 n as the orchestration layer for this automation and I have used the other tools like Google Drive Convert Hub for converting PDF to png.
00:23
I've used Airtable for storing the data and Gmail for sending emails and vision models Gemini and GPT4O.
00:35
So let me show you how the table tables look.
00:39
So I've created two tables.
00:41
One is the COVID reviews which includes all the records that the automation creates.
00:48
So it includes ISBN, book title, author name, status, all the scores percentages.
00:54
It's a pass or fail.
00:56
we have image quality, critical issues, summary of the issues here and corrections suggested by our automation.
01:05
It also has the date and the second table I created is automatic author directory.
01:11
I've created a sample for our tests.
01:15
It includes ISBN, author name and email id.
01:17
it is useful for the automation to pick up the email and send the emails to the authors.
01:26
to show the demo I'm taking these two book covers.
01:31
On the left side we have Benny James SDB which is example for a good cover.
01:36
On the right side we have Ogil Jain which is example for bad cover.
01:39
It has critical issues in the bottom.
01:45
So let's see how it works.
01:50
So I will upload these two images to Google Drive this folder and once I upload will this automation will process it.
02:00
So let me just upload these two.
02:23
So I uploaded them this will trigger the automation.
02:33
You can see that it is running right now.
02:35
Once it completes the run I will receive the emails automatically.
02:45
So looks like I already got one.
02:51
So this is the approved example for approved email it will look like this.
03:01
So your book cover has been approved.
03:03
there's nothing more to do and this is the template I created for pass.
03:10
let's look at the review needed.
03:16
So this is the example for review needed which is Ojal Jain's cover.
03:22
So here it includes everything what issued issues are detected, how to fix the issues and it also has pre submission deadline.
03:33
So here if you can see it is flagging the critical error which is the batch overlap.
03:43
Yeah this is the example for review needed email.
03:55
So let me quickly show you the how each node is working here.
03:59
So this Google Drive trigger is working whenever a new file is uploaded to that folder and it will pick up all the new files and process each one through this loop.
04:11
And we are here, we are extracting the ISBN and metadata of that file.
04:15
We download the file next and then we check whether the file is a PDF or PDF png.
04:22
If it is an image, we directly process it for vision processing.
04:25
But if it's a png, we convert the file first using converthub API and then process the file using our vision models, which is Gemini and GPT4O.
04:37
So once it's processed, it will parse the output.
04:41
Then we look up the email ID in the authors table based on the ISBN.
04:46
And then we create a validation record with all the output we got from this vision model.
04:52
And then we check if it is, pass or review needed.
04:56
So if it is a pass, we send this template.
04:58
If it is review needed, we send this template.
05:01
So that's how it's working right now.
05:04
and if I had more time to improve this, maybe I would, add a second validation check using OpenCV, and also maybe provide the exact markings, where exactly the error has happened, using, markers automatically.
05:20
And then maybe a review dashboard, to manage all the flagged covers.