IDP solution

[random viper]

Are there any other IDP solutions that can handle extracting line items from unstructured documents as well as ABBYY Vantage? I have not come across any other solution that has a feature similar to ABBYY's "continue table from this row".

[final roost]

Maybe try asking in #❔・general-help? There was recent discussion about IDP solutions, though not actually covering feature you mentions. But there were several names IDP solutions mentioned.

[gleaming orbit]

Sounds like a use case for LLMs and Prompt Engineering

[final roost]

Interesting insight... Anyone noticed change in pricing of IDP solutions as LLMs appeared? Or these two actually are not on similar level?

[gleaming orbit]

Traditional IDP solutions don't use LLMs. In my work the best solution we have found is a combination of traditional document understanding and LLM implementation in the case of more unstructured or variable documents

[final roost]

So LLM are simply better or just there are tradeoffs?

[gleaming orbit]

I'm no expert, I just use these tools (the prebuilt ones). LLMs are generally more computationally expensive, but easier to use. LLMs can also suffer from hallucinations, so that's one big downside.

Each has their own benefits, for example:
If you have a form of a consistent structure, it makes sense to use a custom taxonomy to extract the fields you want.
If you have a form of inconsistent structure, you could use an LLM to try and understand and extract the specific data you want. This can be done relatively easily by giving examples etc.

One of the biggest benefits of LLMs is, you no longer need ,000s of items in your dataset to train or use a model, and they have a wide range of capabilities. Whereas before you would've likely have to use multiple different AI models to achieve a task which can be done by LLMs

[stable socket]

I have worked on one of the IDP tool "Datamatics Trucap".
Flow is not that difficult to extract data from unstructured document.
We need to train the different types of documents after that by applying some ontology rules we are able to extract data easily.

/rank