Looking for Idea Feedback - Document keyword categorizer / organizer
The saas is B2B or B2E
Scan a large volume of documents and then provide a search interface with automatically generated categories, key words and descriptions. If this is not just a document discovery/investigation process but perhaps a migration or consolidation from multiple locations then it can make a copy and automatically file the documents organizing by generated keywords/categories and removing duplicates. If there is a set of documents already categorized then the system can learn that categorization, and take non-categorized documents and categorize them.
Use cases would be companies migrating documents from data centers to cloud. Auditing and Accounting teams performing fund or M&A due dilligence, tidying up of archive data once records holds or regulator holds expire..
Tool costs to run 30c an hour and processes 25,000 documents an hour with a 97% true positive and less than 0.1% false positive. True negatives are simply filed / tagged as misc. Documents can be OCR or Images. System can generate images from all Windows, Mac and Google office suites as well as pdf’s
Would be interested in any other ideas for the tool... any negatives, issues, difficulty in finding customers.. any suggestions for target audience outside of financial/accounting enterprise (my domain of expertise) would also be great, you don’t know what you don’t know..
Mostly this is a one-time process, I’d rather convert or have a way to monetize as a subscription.. any ideas/suggestions appreciated
Will you use AI or ML on it? It's an interesting area and can be monetized if AI and ML is being used with deep learning techniques.
Absolutely, its transfer learning on a open source google sota model trained to perform NLP tasks. The final layers where replaced with a hierarchical attention network added to the pipeline which was itself trained on 600,000 generic documents with known categories. Those results i mentioned are based on applying this model to a collection of 60,000 unstructured documents the model had not seen, that is used in production by a legal department and a large financial company.