Dialpad has been granted a patent for a computer-implemented method that enables the automated creation, testing, training, adaptation, and deployment of AI models. The method involves creating a new development version control branch, updating a list of active model package versions, packaging code and trained model artifacts, and merging the version control branch. It also includes running AI model package versions offline from the user site and providing output to the user site based on received live input. GlobalData’s report on Dialpad gives a 360-degree view of the company including its patenting strategy. Buy the report here.
According to GlobalData’s company profile on Dialpad, intelligent contact centers was a key innovation area identified from patents. Dialpad's grant share as of June 2023 was 1%. Grant share is based on the ratio of number of grants to total number of patents.
Automated creation, testing, training, and deployment of ai models
A recently granted patent (Publication Number: US11675581B1) describes a computer-implemented method for managing and deploying Artificial Intelligence (AI) models. The method involves creating a new development version control branch on a main canonical branch of a version control repository for a received new AI model. The method also includes updating a list of active model package versions and packaging the code and trained model artifacts of the new AI model into a versioned model package. Once the versioned model package is approved, it is merged into the main canonical branch of the version control repository.
The method further involves detecting periodic receipt of live input from a user site. In response to this, at least one instance of an AI model package version is run offline from the user site in a corresponding model runner compute instance pool. The received live input is provided to the AI model package version, and the output is exposed to the user site.
Additionally, the method includes running one instance of an AI model package production version and one instance of a corresponding new or updated AI model package non-production version in parallel, in separate model runner compute instance pools, in response to detecting periodic receipt of live input from the user site. Both instances receive the live input and provide output to the user site.
To expose the output to the user site, the method involves storing the output in queryable data storage. The content is periodically retrieved from the queryable data storage and automatically provided to the user site.
Furthermore, the method includes automatically running a declarative directed acyclic graph (DAG) corresponding to the new AI model. The DAG generates a runtime container for the versioned model package based on the new AI model. This runtime container includes generic model running code, extract-transform-load code, and the versioned model package.
Overall, this patented method provides a systematic approach to managing and deploying AI models, allowing for efficient version control, parallel processing of live input, and automatic retrieval and provision of output to user sites.