Botminds v18.7 Release
28 February 2023
Botminds AI Release v18.7 despite coming at the heels of v18.5 is packed with several new capabilities.
The key features and enhancements in this release are:
- Offline mode via RTC
- Datasheet Sync
- Dynamic query pane configuration at view level
- Email Source snapshot
A closer look at the capabilities -
1. Offline mode via Run Time Container (RTC)
This capability is a big leap forward and is significant as it allows prediction models to be available in the offline mode. All microservices are included in a package we call the Botminds AI Run Time Container (RTC) - which allows it to run in completely offline mode. The RTC can be downloaded and run in any environment and system, without being dependent on the Azure environment.
This can later be synced to the online environment, if required.
This is useful for organizations that want to maintain the security and privacy of document data, and run their processes on an Intranet, for instance; and not expose the data to the external environment.
There are multiple other advantages of this capability – it improves the efficiency of the model, enhances system performance and reduces the infrastructure cost to run an application.
A significant aspect for clients is that, with this capability, the Botminds AI platform is available in the On-Premise model (apart from the cloud-based model).
2. Datasheet sync
Previously, only one file could be ingested at a time in the Botminds AI platform. The current release has added the ability to ingest a folder with multiple files. All the files are ingested into a single datasheet. The datasheet view includes columns across the files. The columns across the files may be different – if a file does not have a particular field, the field value is blank.
Earlier, the datasheet (any excel/sheet/SQL data etc.) was ingested as a one-time import, and each time there was a change in the source data, it would need to be imported again. With this enhancement, any required sync frequency in multiples of minutes/hour/day/week etc. can be set up for the data sheet. The source (db/google sheet etc.) is looked up periodically and changes, if any, are updated in the datasheet. Based on business need, near real-time sync (10 min or less) can also be set up; without repeated ingestion of the same sheet if there are data updates in the source file.
3. Dynamic Query pane configuration at View level
The search query pane in the platform used to be static at a project level. This was a limitation, since not all fields are relevant across different views – however, only one search query could be saved per project common across Views.
With this release, users now have the ability to save search query filters (for each column) at a saved View-level. Queries could differ based on users and specific needs, this dynamic query pane ability greatly improves the ease of use and productivity for users. Users can quickly pull up the view and use the query associated with it for their analysis.
4. Email Source snapshot
This enhancement provides the option to have Email as a screenshot image that is ingested (the image is stored in the cloud at the end of ingestion and provided as a URL in JSON format to end users). An option to download the screenshot image is also provided in the Botminds AI platform. The screenshot image URL is added in the API along with other document information such as Title, Document URL, Summary etc.
This is a small but important enhancement to the document types ingested by the platform. One of our clients requires email snapshots as part of document proof in their process – and this enables them to download the verification/proof.
The v18.7 release also includes the following improvements to the platform -
- Structured Document Ingestion
- Data Sheet Support for JSON & API
- Consolidated lookup for document set at document listing level
- Prediction explanation
- Improved ingestion speed and stability
1. Structured Document Ingestion
Annotations can now be mapped automatically in the summary columns for ingested datasheet (Auto-annotation). An output document can be rendered leveraging the consolidated column set – this allows easy modification of output formats, layouts – without having an actual physical file. Analytics can be built on top of this, and model training can be enhanced. Unstructured document data combined with structured data input make the predictive intelligence output more accurate & insightful.
This feature simplifies the overall ingestion & annotation process in the platform.
2. Data Sheet Support for JSON & API
Key-value pairs are used by developers and users to configure data. With this release, key-value pair files can also be imported. The API response comes as JSON files, which can now be saved and viewed as documents, and have prediction models run on it this well.
3. Consolidated lookup for document set at document listing level
Derivation logic can now be applied across related data points across documents in a document set & the derived or calculated data point can is displayed in the side-by-side view of the Botminds AI platform. Previously, the derivation logic could be applied only at an individual document level. For instance, the CGPA score may be available for a student record in 3 documents in the document set – the total or the average of the score (as required) can be derived and displayed in the output view.
4. Prediction explanation
This capability is an important one. With this release, users can see the list of input data points or influences in the output of the prediction model. The influences could be positive or negative. This is based on the concept of SHAP values (SHapley Additive exPlanations) - which is a method to increase transparency and understanding of machine learning models. SHAP shows the contribution (or weightage) of each attribute on the prediction of the model, but it does not evaluate the quality of the prediction itself.
5. Improved ingestion speed and ability
Changes made under the hood to improve ingestion performance and provide a smoother experience for users.