Seeking Solutions: Preserving Table Structure in JSON Output with Adobe PDF Extract API for RAG App

[sai Amith37951804r1ly]

Hello all! Iused adobe extract pdf API service to parse a pdf. Pdf and output JSON is attached to this message.  I believe the Json output doesn't preserve the table structure. If I pass this data to an LLM, it is not able to answer relevant questions about this data as the table structure is not preserved. How should I go about this? I want to use adobe API to build a RAG application. Is there a way to preserve the table structure within the Json file? for example, I need outputs such as like this:

{

"Input (DC)":

{ "MVPS 4000-S2": null

, "MVPS 4200-S2": null

},

"Available inverters": {

"MVPS 4000-S2": "1 x SCS 3450 UP or 1 x SCS 3450 UP-XT",

"MVPS 4200-S2": "1 x SCS 3600 UP or 1 x SCS 3600 UP-XT"

},

"Max. input voltage": {

"MVPS 4000-S2": "1500 V",

"MVPS 4200-S2": "1500 V"

},

"Number of DC inputs": {

"MVPS 4000-S2": "dependent on the selected inverters",

"MVPS 4200-S2": null

},

"Integrated zone monitoring": {

"MVPS 4000-S2": "○",

"MVPS 4200-S2": null

},

"Available DC fuse sizes (per input)": {

"MVPS 4000-S2": "200 A, 250 A, 315 A, 350 A, 400 A, 450 A, 500 A",

"MVPS 4200-S2": null

},

I know it can also generate csv, but the csv doesnt have any other information that might be present in the pdf. 

[Joel Geraci · Accepted solution]

I generally post-process the JSON from extract to create a Markdown file. When I hit a table, I read past it, read in the .CSV as a Markdown table, then contuinue with the JSON. It works great. I have some Node.JS code I can share if you like. 

[sai Amith37951804r1ly]

Hi Joel,

Thanks a lot for the reply! Yes, would really help if you can share your Node.JS code.

[Joel Geraci]

It's in a private git repo. If you are comfortable doing so, send me a private message with your github ID and I'll add you as a collaborator. I eventually plan on making it opensource once I'm past the work-in-progress. 

[sai Amith37951804r1ly]

Thanks Joel! I just sent you a personal message that has my Github ID. 

[sean_2462]

were you able to get the table structure into the json? If you have the code please let me know, thanks

[sean_2462]

Hi Joel

Were you able to make it open source? If so can I also get acces to the code?

Thanks

[k_1671]

I have tried to convert directly PDF to JSON via Adobe API. I just wanted to test it because of this mad pricing policy. However the result wasn't exact at all. I had complex tables with merged cells. The solution that worked for me was: 

1. convert PDF to DOCX via eg CloudConvert (they provide much better pricing with credits)

2. then convert DOCX to JSON.

That worked perfectly!