Quantiphi, Inc

A method for automatic visual workflow model generation and management is disclosed that utilizes multimodal inputs and user feedback. The method further comprises receiving, through a processor, descriptions using an advanced AI model, generating elaborate plans that visually organize sequential tasks. User feedback via natural language on these plans refines them, establishing connections between detailed plans and numerous sub-skills. The processor constructs a directed acyclic graph (DAG) visualizing sub-skill execution order based on the established mapping, culminating in an executable workflow model. The method further comprises seamlessly translating user descriptions into detailed plans, refine them iteratively, and generate an executable workflow model, all driven by user interactions and advanced AI techniques supporting natural language understanding and generation.

Methods, systems and computer program products for generating standardized structured data from unstructured and semi-structured images of document pages is disclosed. The embodiments include a training framework where, boundaries of one or more instances of a first and a second set-of-fields are detected, from images of document pages and tagged using unique labels. Individual fields within the set-of-fields are identified and associated with each instance and the unique labels, to generate large number of synthetically labelled documents. A neural network model is trained using the original document image and the large number of generated synthetically labelled documents. An inference framework receives as input scanned images of unstructured and semi-structured document pages. Custom object recognition module identifies different set-of-fields, and an OCR module recognizes the text from the input images. The outputs of these modules are stitched together to create a standardized structured data.

A method and system for multi-table question answering is provided herein. The method includes receiving a complex question from a user. The method may further include decomposing the complex question into one or more sub-questions using one or more of but not limited to coreference resolution, dependency parsing, chunking and prompt techniques. The method may further include generating code scripts for each sub-question using a Large Language Model (LLM). The method may further include executing the code scripts to retrieve relevant structured data from a plurality of tables. The method may further include generating an answer to the complex question based on the retrieved structured data.

Disclosed is system for enabling generation of industry-tailored virtual assistants. System comprises server(s) and data repository communicably coupled to server(s), wherein server(s) is configured to: receive raw conversation data from at least one data source, transform raw conversation data to generate processed conversation data, generate taxonomy(-ies) of industry using processed conversation data, label processed conversation data to generate labelled conversation data, create relationships between entities in labelled conversation data and taxonomy(-ies), generate dataset pertaining to digital warehouse that is usable to generate industry-tailored virtual assistants, and provide digital warehouse using dataset on digital portal that is accessible by target device(s) wherein digital portal is hosted by server(s), wherein digital warehouse comprises listings for set of items including one or more of: intent(s), skill(s), virtual assistant template(s), pre-built and pre-trained virtual assistant(s).

A validation system enables concurrent visual validation of multiple electronic documents. A processor generates a custom user interface (UI) framework comprising two sections. The first section displays document previews, while the second holds extracted data, including identifiers for document and page IDs, along with entity positions and associated values across the documents. Upon user input on a specific document entity in the second section, the validation system concurrently loads document previews in the first section, displaying corresponding data values from the multiple documents. The updated visualization allows for validation and review operations across the documents, utilizing the received user input and loaded information.

Disclosed is a method comprising receiving user query (UQ) indicative of task ( 202 ) associated with retrieval of data ( 204 ) from database ( 206 ) comprising structured and unstructured data sources (SAUDS) ( 208 ) and generating relevant response (RR) ( 210 ) for UQ; generating task plan ( 220 ) for determining embedding ( 224 ) for task, performing retrieval task ( 226 ) for identifying relevant SAUDS ( 228 ), generating relevance scores ( 230 ), and when relevance scores indicate unstructured pipeline ( 232 ): triggering unstructured data query engine (UDQE) ( 234, 306 ) for retrieving data, and generating RR, or when relevance scores indicate structured pipeline ( 236 ): triggering structured data query engine (SDQE) ( 238, 308 ) for retrieving data, and generating RR, or when relevance scores indicate hybrid pipeline ( 240 ): triggering SDQE for retrieving first portion of data (FPD) ( 242 ), transforming received UQ and FPD, triggering UDQE for retrieving second portion of data (SPD) ( 246 ), and combining FPD and SPD for generating RR.

A method and system for training physics-informed operators with dual hypernetwork module and low-rank domain decomposition. The method includes identifying availability of information related to system behavior and geometry complexity. The method further includes dynamically determining training strategy for the PINO based on the identification. The training strategy includes at least one of a soft domain decomposition process or a hard domain decomposition process. The method includes determining whether a size of a hypernetwork output layer exceeds a first predefined threshold. The method includes generating a set of training points if the size does not exceed the first predefined threshold. The method includes iteratively training the PINO using the determined strategy based on the set of training points until an error falls below a second predefined threshold. The method includes storing a trained model of the PINO with an associated set of parameters when the error falls below the second predefined threshold.

The present disclosure provides a method for recommendation of suitable assets using Large Language Model (LLM), method comprising receiving descriptors for each asset from amongst plurality of assets; generating embeddings for each asset from amongst plurality of assets, by employing LLM; creating database of assets by storing generated embeddings for each asset from amongst plurality of assets in database; receiving user query pertaining to request for recommendation to solve enterprise problem; generating cosine similarity scores, wherein each cosine similarity score is generated between user query and corresponding generated embedding for given asset stored in database of assets; identifying predefined number of similar assets from amongst database of assets, based on highest cosine similarity scores; constructing optimal prompt based on identified predefined number of similar assets and user query; and prompting LLM, using constructed optimal prompt for generating response as recommendation of identified predefined number of similar assets as suitable assets for solving user query.

A method and system for multimodal input processing for a virtual agent is provided herein. The method comprises obtaining a multimodal input by the virtual agent from a user. The method further comprises identifying a plurality of principal entities within the multimodal input. The method further comprises extracting information about each entity of the plurality of principal entities. Further, the method comprises generating a response based on the extracted information.

A method for processing one or more electronic documents for enhanced search includes defining a bounding box around each key and value of key-value pairs in a first schema file, tagging coordinates of a key corresponding to a first bounding box and coordinates of a value corresponding to a second bounding box in the first schema file. Furthermore, obtaining a first inference file, detecting coordinates of a key corresponding to a third bounding box, and determining coordinates of a fourth bounding box and the value of the first inference file that are determined by applying a normalization operation. Thereafter, extracting value encompassed by the fourth bounding box of the first inference file and automatically creating a searchable index of the first inference file with searchable key-value pairs. The method achieves an efficient and accurate clustering of data items with an accurate, meaningful, and formal objective function.