Advanced electronics datasheets parsing system

[Plinsboorg]

Hi all,

I'm writing to show you the project I'm working on.
This is an app that is parsing component datasheet, and let's LLM use the data with the ability to always cross check the result with the original document. Right now we've only implemented parts comparisons and we're working on expanding the features

Here is me describing it on video:
https://youtu.be/vhfAR3Q6SBM
We're still developing the app, but I'm looking for as much feedback as I can get to steer the development in the right direction

This may be useful for your work too. Let me know if you want to get premium access

[tggzzz]

Hi all,

I'm writing to show you the project I'm working on.
This is an app that is parsing component datasheet, and let's LLM use the data with the ability to always cross check the result with the original document. Right now we've only implemented parts comparisons and we're working on expanding the features

Here is me describing it on video:
https://youtu.be/vhfAR3Q6SBM
We're still developing the app, but I'm looking for as much feedback as I can get to steer the development in the right direction

This may be useful for your work too. Let me know if you want to get premium access

What benefit would there be to me if I spent 18 minutes of my remaining life looking at your video? Hint: I can speed read at least 10 times faster than you can talk.

What benefit would there be to me if I used your app? Your answer should take less than 30s of my time

[Plinsboorg]

Fair point.
Here is a written article: https://wp.plinsburg.tech/2024/11/13/factframe-project-presentation/

Video is describing the UI.
But if I need to put it in text, then basically you upload the datasheet to the website, it takes several minutes to parse it.
At this point you can already download the contents of the datasheet in excel format (with correctly recognized merged cells). This may be useful by itself if you want to get tables from the datasheet into your internal documentation or footprint generator or what have you.
Then the tool offers possibility to compare the components. This feature is geared towards corporate use where they need to make comparison documentation before making a decision that any given alternative part can be used to replace the original.
But I used this to demonstrate the idea of FactFrame, which is a popup window showing a screenshot of the section of the original datasheet where the data has been taken from.

When I was making worst case analysis calculations at my last job I basically had to do the same. Whenever I declare a variable I need to put a screenshot of the datasheet where it's taken from nearby so the value can be checked.

I think having such checking mechanism is useful for LLM use in electronics domain. If LLMs output can be fact checked with a single click, I think it would be good for the adoption of LLMs in our domain. Especially in areas of generating documentation, making libraries, and generally taking info from the datasheets.

I'm looking for feedback on my ideas, and most of all for potential users. I want to work on the features that would actually be useful to people.

[tggzzz]

OK; that's sufficient.

You write...

"EEs have developed a set of practices over the years on how to detect and avoid design mistakes". "FactFrame technology can be usefull for any industries where the cost of mistake is high and the knowlege is mostly stored in the “source of truth” documtents."
Yes indeed. One of my techniques is to distrust LLMs, since at best they merely group words that are probably found in similar contexts, and at worst it is inevitable that they hallucinate.

"This tool uses special PDF parsing and LLM prompting techniques to allow LLM to remember where exactly in the document the information is taken from."
That is of no benefit to me; data sheets follow conventions and hence I know where the information will be found in a datasheet.

"It allows to speedup “paperwork production” by LLM while still leaving humans full control and visibility over what had been generated."
You need to justify that assertion. Current experience is that LLMs require extra time checking that their output is correct (and not a hallucination), thus negating possible savings.

As a separate issue outside the scope of your project, data sheets do contain mistakes, and sometimes they represent what can be tested/guaranteed rather than device performance.

Good luck with your project. Be aware that engineers will "kick the tyres" of anything and everything before deciding to use it. Hence having solid repeatable evidence of benefits is very useful.