Added a Security & Privacy page to the showcase doc
Can also be found in top nav:
The Privacy Policy has been updated to reflect that I now store names and emails
@Christiaan_Huizer has provided a helpful perspective for the development of this pack, and he just released a nice blog about it 
This pack is incredible! I have just spent a bit of time testing this pack with a will and a report on a mining company. Fabulous results - saved me a ton of time finding out what I needed to know. Check it out. Wizardry!
The server is now using multiprocessing, giving us ~3x speed for bigger documents!
A 66 page PDF used to take 43 seconds - It’s now done in 14 seconds
This will allow processing of even larger documents within the 100-second timeout
Keep in mind, if you expect more than:-
ReadTextFile() on the Text File output, worst case Coda will automatically cancel your formula if you try to display it directly
You can now optionally send the scanned text to an AI along with a prompt!
Example: Summarize a 66 page PDF
Scan()// Get text like normal
Scan(PDF)
// Send the text to the default GPT-4o mini
Scan(PDF, prompt: "Summarize this please")
// Use the smarter but costly GPT-4o
Scan(PDF, prompt: "Summarize this please", engine: "gpt-4o")
You can see separate AI info for each request:
{"ai": "gpt-4o-mini", "pages": 2, "ai_cost": 1, "file_count": 2}
I’m excited to announce a much-requested feature!
Automatically create rows from images and PDFs in any structure you choose
Here’s a showcase video! (5 mins)
I’ve worked hard on a general solution. For example, I asked it to process my profile picture:
Or some shapes with no prompt:
The Scan() formula now supports more engine options, like engine: "gpt-4o-mini", which allows AI-based OCR. This means it can capture extra details like colors, text positions, and more.
Path column to Requests sync tableEngines sync tableimpressive work and contribution, extracting a pdf and getting intelligent outcomes was never so easy!
Fantastic ! Thank you so much!
Do you have a suggestion for best practice for designing a system that utilises this pack for the following scenario:
Extracting summary and row data from invoices from 20+ suppliers, where the formatting of invoices varies from supplier to supplier?
To be a bit more specific about me question, do you have a suggestion for how to scale up the design you demonstrated in this youtube video,
https://www.youtube.com/watch?v=oe3sKeo5B0I&t=4s (Convert PDFs to Coda.io Rows Instantly with OCR Pack!)
such that a wide range of invoice formats can be processed?
Thank you for creating this pack, and for the fantastic explainers, they are making the pack so much more accessible for me!
Hi Sam thanks for your question!
That setup in the video should be able to handle it!
I’d recommend considering adding some columns to the Invoice table such as
as well as any other data points that generally occur within your invoices, otherwise the AI will probably put it in the Other Info column if you have that one
Imagine it’s a person you’re sending a message to to fill out your Coda rows, is there enough information to understand what values each cell should have, and how many rows per table?
If not then make your table names and column names more descriptive or simply give additional instructions with the prompt parameter
I’ve worked hard to accomplish a general dynamic solution, it’s an involved process under the hood, but for the user I’ve hoped to make it as intuitive as possible. If the AI has issues at any point then it will abort the process and create a nice error message for the user, hopefully informing what actions they can take to make it work
Fantastic!
You’ve given me plenty of info to keep going on with. I’ll report back with any progress updates.
Here’s a flowchart for my OCR and PDF packs, which I understand are confusingly similar 
And thanks to @Christiaan_Huizer for his continuous support and latest post!
This is SO helpful. Thank you for taking the time to put it together.
For anyone who has used this pack in the past and gotten mediocre OCR results – I highly recommend trying again, and be sure to try one of the new models. I’m finding great results with both gpt-4o and gpt-4o-mini. Totally amazing.
Thank you @Rickard_Abraham!
Indeed it is a fantastic pack with great capabilities, I wrote about it on LinkedIn.
This pack means no more manual data entry, no more tedious searches, and no more missed insights.
I’m super honored you tried my pack
Thanks a lot for the feedback!
A few small suggestions:
Thanks!
This is great! I’m always looking to improve 
That makes sense. I’ve tried my best to plan for the future by making the default engine “Balanced”, as new models will continously appear this allows me to accomodate for that. It’s a very interesting problem though, as all models vary in Cost, Speed, Quality, and I/O.
For the cost and speed I’ve done some linear regression to get estimates
Oh, making Scan an action would have been much better, I honestly didn’t know actions could return data to the user, would you look at that
Alright got a nice little update!
gpt-4o now points to an updated model which is ~50% cheaperScan() formula is now an actiongpt-4o-mini if a prompt is providedg-doc-ai-v2 if no prompt is providedThanks again for the feedback @shishir
Changing Scan() to an action is technically a breaking change, but it will only affect workflows where it’s been used outside a button.
Using Scan() as a formula is strongly discouraged as it will consume your credits with every automatic recalculation.
And also, as this pack isn’t seeing too much traffic, this felt like the best path forward
Ah the breaking change did trip me up a bit - but agreed it is much better for this to be an action instead of a formula. Thanks for the updates!