HI Team,
We are Using AI Skill Extract Data from Documents (IDP).
At starting we trained a model with Driving License with around 25 Documents and its recall is 96% and it took around 7-8 minutes to extract the Data
then trained with 50 and its recall is 90% and its taking 5-6 minutes lastly i tried with 75 documents recall is 83% taking 5-6 minutes to Extract data.
i have requirement to extract data with in less than 30 seconds or max of 45 seconds.
How should i achieve this extraction of data from documents with in seconds
Discussion posts and replies are publicly visible
Where is this strict timing coming from? I mean, typically, in a mostly automated process, people are not waiting for such things.
Actually we had a Requirement When the User Uploaded their Driving License with in few seconds in the next , we need to extract and show them in interface.
ok leave about the timing , but can we reduce the extraction in any chance, is there any way to do so that the extraction time can be reduced ?
I understand.
AFAIK there is nothing you can configure or change to make this go faster. I suggest to contact Appian to discuss this use case.
Okay Thank you Stefan
Hi Vinay, AI Skill Extract Data from Documents (IDP) usually takes a lot of time to extract, no matter how many times you train it. I would suggest you to use "AWS Textract" integration to extract text from doc's, It takes less than a min. Only thing that you need to take care of document size limit since AWS textract doesn't take more than 5 MB. So before sending document to AWS textract integration, use "split pdf by number" smart service and send one by one using MNI to AWS Tectract.
Hi Vinay,
I think there is no chance for us to provide particular timing to the AI skill to extract the data. I observed some scenarios likeWhen you retest the document which is already extracted once, It will take less time to extract the data.For a new document based on the data in document the time may variesI think more advancements to be expected in AI skill in up coming releases regarding this.
Thank you