The image shows the logical architecture consisting of a Compute VM with the
following components:
- Router: User requests go to the Router which identifies the
request type (SQL generation and data retrieval or data analysis) and directs it to
the appropriate processing pipeline. The component is based on LLM with a
specialized prompt.
- Schema Manager: Manages all metadata related to the database
schema. It plays a crucial role in providing, for each NL request, the list of
tables (“restricted schema”) to be used for SQL generation and all associated
metadata. It leverages a 23ai database and Semantic Search within the database, to
find the tables relevant for the user NL request.
- SQL Generator: Translates the NL request into an executable SQL
statement on the connected "Data" database. It uses an ensemble of LLM models
to enhance accuracy and increase the success rate of correctly generated SQL
statements. SQL Generator is connected from the Cache by a unidirectional arrow and
connects to SQL Executor by a unidirectional arrow.
- SQL Cache: Stores NL instructions (and their SQL equivalents) for
the most frequent requests.
- SQL Executor: Enables syntax validation of the generated SQL
statement and executes the instruction and sends back to the user.
- AI Data Analyzer: Analyzes stored conversational data to
generate responses for analysis requests or report generation.
Data is displayed on the Dashboard from the SQL Agent Monitor.
