Abstract: In some embodiments, a multimodal computing system receives a query and identifies, from source documents, text passages and images that are relevant to the query. The multimodal computing system accesses a multimodal question-answering model that includes a textual stream of language models and a visual stream of language models. Each of the textual stream and the visual stream contains a set of transformer-based models and each transformer-based model includes a cross-attention layer using data generated by both the textual stream and visual stream of language models as an input. The multimodal computing system identifies text relevant to the query by applying the textual stream to the text passages and computes, using the visual stream, relevance scores of the images to the query, respectively. The multimodal computing system further generates a response to the query by including the text and/or an image according to the relevance scores.

Abstract: Systems and methods for disambiguating company profiles are disclosed. The system builds a database of candidate companies with timed attributes. The system further ingests timed company metadata using a PostgreSQL database. The system disambiguates location and geocoding by matching text patterns and cross-referencing one or more identified location components against one or more geocode databases and classifies and disambiguates company name component from a company name associated with the company using a conditional random field (CRF) model. Further, the system disambiguates employee attributes using a Latent Dirichlet Allocation (LDA) topic model algorithm and train a tree model for pre-selection of candidate companies for the company. The system trains a similarity model for comparison of the candidate companies and in response to a determination that two given candidate companies are same merge the company profiles associated with the given candidate companies.

Abstract: Disclosed embodiments include a method performed by server computer(s). The method includes receiving a query and defining a query plan based on the received query. The query plan refers to datasets contained in data sources. The method further includes determining that the received query can be accelerated based on an optimized data structure contained in a memory, where the optimized data structure is derived from a dataset referred to in the query plan. The method further includes modifying the query plan to include the optimized data structure, and executing the modified query plan to obtain query results that satisfy the received query by reading the optimized data structure in lieu of reading at least some data from the data sources.

Abstract: Systems and methods for rapid importation of data including temporally tracked object recognition. One of the methods includes receiving datasets each indicating information associated with one or more objects. Information indicating unique identifying information associated with the objects is accessed, and an updated dataset joining information from datasets that is associated with each object is generated. The updated dataset is maintained to include most recent versions of each of the datasets, with one or more datasets being replaced with more recent versions, and with one or more other datasets being propagated to be the most recent versions. Queries received from clients are responded to, with the queries indicating requests for specific information related to objects.