Summary
How can an agent team query tables and documents in Oracle Autonomous AI Database 26ai?
The demonstration uses Oracle AI Accelerator to prepare two sources: a debt-collection dataset in CSV format and a public financial statement from Banco de Chile in PDF format. A group and user are created, the modules are enabled, and the column dictionary and AI profiles are reviewed.
Select AI queries the requested amount grouped by status. Select AI with RAG retrieves information from the PDF. The demonstration then reviews the team, agent, task, and SQL and RAG tools in DBMS_CLOUD_AI_AGENT, and tests a query using that workflow inside the database.
The closing section discusses possible future automation using functions, notifications, and other tools, and introduces Oracle AI Data Platform. These possibilities are presented as extensions of the example.
Start
The demonstration introduces the Oracle AI Accelerator update for the AI capabilities of Oracle Autonomous AI Database 26ai. The walkthrough combines queries over structured data, document retrieval, and agents.
Select AI and natural-language queries
Select AI lets users ask questions in natural language and generate SQL queries over tables. This capability is reviewed as a starting point before adding documents and an agent workflow.
Select AI with RAG and documents
The video explains information retrieval from documents stored in a bucket, such as PDF, TXT, or HTML files. Vectorization enables content searches to answer questions and prepare summaries; the later example uses a PDF.
Agents, tasks, tools, and teams with DBMS_CLOUD_AI_AGENT
The video introduces DBMS_CLOUD_AI_AGENT and the relationship between teams, agents, tasks, and tools. It describes sequential execution to orchestrate the tools available in this example.
Creating the group and user
An administrator creates the SH group and a new demonstration user. The user details are completed and the group is assigned before enabling the modules the user will need.
Enabling Select AI, RAG, and agents
Select AI, Select AI with RAG, and the agent module are enabled for the new user. The agent is prepared to interact with the two example tools; the configuration is then saved and a session is started with that user.
Uploading the dataset and data dictionary
A debt-collection dataset in CSV format is imported into the knowledge base. The column dictionary is highlighted: its descriptions provide context for interpreting the fields. The upload passes through a bucket before the data is prepared for Select AI.
AI profiles and model configuration
The video explains that Select AI and Select AI with RAG use AI profiles to define model access and parameters. The dataset upload is confirmed before adding the second source.
Uploading the PDF for RAG
A PDF document is added, its description is completed, and the preparation is allowed to finish. The workflow uses the bucket and Select AI with RAG to make the document content available.
SQL queries over the debt-collection dataset
The table and its field comments are reviewed in the Select AI module. A question asks for the requested amount grouped by status. The demonstration notes that response time depends on the resources allocated to the database.
Querying a financial statement in PDF
A query is made against the public Banco de Chile financial statement uploaded earlier. Select AI with RAG retrieves information from the vector database and presents an answer based on the document.
Reviewing the team, agent, and tools
The team created during user configuration is identified and its name is used to run the workflow. The agent, its task, and the two tools available to query structured data and documents are reviewed.
Testing the agent workflow
A question is submitted to the agent workflow inside the database. The configured task orchestrates the available tools to handle the query; chat is used as the demonstration interface.
Automation possibilities and other tools
Possible extensions are discussed: functions that change data when a condition is met, web browsing, and notifications by email or Slack. These are automation possibilities discussed during the closing section; their implementation is not demonstrated in this test.
Oracle AI Data Platform overview
The video gives an overview of Oracle AI Data Platform and its relationship with the autonomous database, agents, and other information sources. Connections to buckets and external sources are discussed within a broader data and AI ecosystem.
Next steps and closing
The presenter announces that more information about these integrations will be shared and closes the demonstration.