Data Interaction
Data interaction refers to the mechanisms by which intelligent systems, particularly llms and agents, query, interpret, and manipulate structured or unstructured data sources. This concept encompasses direct API calls, semantic layer queries, and tool-augmented retrieval patterns.
Core Principles & Patterns
- Semantic Querying: Translating natural language into executable queries (e.g., SQL, LookML) via intermediate representations.
- Tool Use: Leveraging external tools to bridge the gap between model context windows and live data stores.
- Agent Orchestration: Coordinating multiple steps—retrieval, validation, execution—to ensure accurate data responses.
Integration Architectures
Agent Development Kit (ADK) & MCP
Recent implementations utilize the Agent Development Kit alongside the model-context-protocol to standardize agent-to-data interactions:
- MCP Toolbox: Provides standardized connectors for databases and BI platforms, allowing agents to discover available data schemas dynamically.
- ADK Configuration: Enables modular agent design where specific tools (like Looker connectors) are registered as capabilities within the agent’s loop.
Case Study: Looker Integration
A practical implementation involves configuring an LLM agent to interact with Looker dashboards and datasets:
- Setup: Uses ADK to define the agent structure and MCP to expose Looker’s data endpoints.
- Workflow: The agent receives a user query, identifies necessary metrics via MCP schema introspection, executes the query through the configured adapter, and synthesizes the results.
- Reference Implementation: See Configuring an LLM Agent for Looker Data Interaction Using ADK and MCP for detailed configuration steps involving Gemini 2.5 Flash.