Skip to content

v2.0.0

Latest
Compare
Choose a tag to compare
@5Y5TEM 5Y5TEM released this 28 Aug 16:22
· 54 commits to main since this release
283d2a5

Release Notes - Open Data QnA v2.0.0

This major release brings significant improvements and new features to Open Data QnA.

Multi turn capabilities

Ability to interact back and forth with the database in a context. Initial v1 was established with a single turn query. In this release, we have created a multi turn architecture that saves the session info, previous query information and can answer accordingly. For more information on the architecture: link

Table Grouping

Initial v1 was tied to single dataset processing and all the tables under this dataset. In reality, users most likely want to restrict the tables and add other datasets if needed. This table grouping provides a way for users to be able to define their scope

Data Sampling

We provide a sampling of data values in a column to provide contextual information to the SQL Generation agent. For this, top 5 values are retrieved for every column in the specified tables.
This information is aggregated and stored back into the vector store, and is retrieved during the retrieval process.

Data summarization

In the initial V1 release, the results were in tabular format. With this release , we provide summarized answers in a natural language format that can be integrated into a chatbot. User does have an option to still get the tabular and visualized results based on their settings.

Resolving ambiguities

The multi-turn approach helps to resolve ambiguities in the questions, by allowing the user to provide follow-up questions and clarifications.

Furthermore, it is possible to provide additional context in the instruction prompt to let the LLM resolve ambiguities before triggering the pipeline. This can be achieved with the help of a LLM router added as a first layer before the Open Data QnA pipeline.
These clarification questions can help provide more context to the SQL creation.

Ambiguities can be categorized into semantic, application, business and database context. With this release we look for semantic and business level context and resolve such ambiguities through the chat interface.

UX through Flutter and Streamlit

In addition to the AngularJS, we have added support through Flutter as part of the release which can be found under the front end code folder.
Furthermore, to enable more efficient development, we have added support for streamlit, so users can quickly iterate and test in a dev frontend before deploying to Angular or Flutter.