AI BriefingArchitecture and Design Principles of Shippy, a Maritime AI Agent
Key point
Introduces the architecture of Shippy, a maritime AI agent for high-stakes decision-making, and its strategy for ensuring reliability.
Details
This covers the technical structure of Shippy, an AI agent designed for environments like the maritime domain where errors can lead to major risks. Beyond simple model performance, Shippy focuses on securing system Reliability.
The agent's structure is defined by three main elements:
- Soul: The System Prompt that sets the persona and behavioral boundaries. Boundaries are set through prompts rather than tuning, making the system easy to audit and modify.
- Skills: Functional units that handle specific requests. Managed as markdown files following the Agent-skills spec, including functions such as querying the Skylight API, looking up EEZ boundaries, and interpreting vessel trajectories.
- Config: Runtime environment settings. Includes LLM and runtime configuration such as the OpenClaw framework and Claude 3.5 Sonnet (the mention of Opus in the text appears to be an error, but contextually refers to the latest model).
To compensate for the non-deterministic nature of LLMs, Shippy leverages Deterministic tools. For example, when asked about fishing activity in a specific sea area, instead of having the model guess coordinates, it first fetches the exact EEZ boundary polygon via API and then queries the data.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.