Systems and methods for ai-based document management and automated electronic submission template and resource form completion

A method performed by one or more computing devices, the method comprising:

• receiving a plurality of electronic files associated with an electronic Submission Template and Resource (eSTAR) form;
• extracting, using a pre-trained natural language processing (NLP) model, semantic information from the plurality of electronic files;
• generating, based on the semantic information, a plurality of high-dimensional embeddings;
• segmenting, based on the extracted semantic information, the electronic files into a plurality of content slices using a sliding window method, wherein each content slice of the plurality of content slices is associated with a corresponding high-dimensional embedding of the plurality of high-dimensional embeddings;
• generating a vector database comprising a plurality of vectors, the corresponding high-dimensional embeddings, and one or more metadata tags, wherein the plurality of vectors represent the plurality of electronic files;
• receiving an indication of one or more sections of the eSTAR form, wherein the indication comprises a selection of the one or more sections on a user interface displaying the eSTAR form;
• converting, based on the pre-trained NLP model, the indication of the one or more sections of the eSTAR form into one or more query embeddings;
• determining a set of content slices of the plurality of content slices based on an optimized similarity search of the vector database for the one or more query embeddings, wherein the optimized similarity search comprises using machine learning to determine a degree of semantic similarity between the one or more query embeddings and one or more high dimensional embeddings stored within the vector database;
• converting the set of content slices into a data format compatible with the eSTAR form;
• and
• inserting each of the set of content slices into the user interface using an API based on a mapping to the one or more sections.

A system comprising:

• one or more processors;
• and
• memory coupled with the one or more processors, the memory storing executable instructions that when executed by the one or more processors cause the one or more processors to effectuate operations comprising:
• receiving a plurality of electronic files associated with an electronic Submission Template and Resource (eSTAR) form;
• extracting, using a pre-trained natural language processing (NLP) model, semantic information from the plurality of electronic files;
• generating, based on the semantic information, a plurality of high-dimensional embeddings;
• segmenting, based on the extracted semantic information, the electronic files into a plurality of content slices using a sliding window method, wherein each content slice of the plurality of content slices is associated with a corresponding high-dimensional embedding of the plurality of high-dimensional embeddings;
• generating a vector database comprising a plurality of vectors, the corresponding high-dimensional embeddings, and one or more metadata tags, wherein the plurality of vectors represent the plurality of electronic files;
• receiving an indication of one or more sections of the eSTAR form, wherein the indication comprises a selection of the one or more sections on a user interface displaying the eSTAR form;
• converting, based on the pre-trained NLP model, the indication of the one or more sections of the eSTAR form into one or more query embeddings;
• determining a set of content slices of the plurality of content slices based on an optimized similarity search of the vector database for the one or more query embeddings, wherein the optimized similarity search comprises using machine learning to determine a degree of semantic similarity between the one or more query embeddings and one or more high dimensional embeddings stored within the vector database;
• converting the set of content slices into a data format compatible with the eSTAR form;
• and
• inserting each of the set of content slices into the user interface using an API based on a mapping to the one or more sections.