A New Paradigm for Deploying ChatBI with Harness and Skill

As large‑language‑model (LLM) capabilities become stronger, the underlying data‑infrastructure can no longer be ignored. Traditional BI products suffer from inconsistent metric definitions, difficult data extraction, and manual attribution, problems that persisted because AI’s semantic understanding and reasoning were not yet mature. The recent maturity of LLMs creates a turning point.

Based on this insight, Ctrip built a practical ChatBI solution. The architecture features a Multi‑Agent collaboration framework where each sub‑agent has a dedicated responsibility, a memory‑management layer that separates short‑term conversational context from long‑term user preferences, and a tool‑orchestration layer powered by Harness combined with Skill management to encapsulate data‑retrieval and attribution functions in a standardized way. For the LLM backend, the team evaluated Claude SDK against Alibaba’s Agent Scope and selected the most suitable option for their scenario.

At the foundation, Ctrip constructed a metric engine and a topic‑governance system that organizes knowledge documents and attribution capabilities. Specific techniques are provided for core challenges such as metric identification and dimension recognition. Quality assurance is handled through accuracy evaluation, automated testing, an active clarification mechanism, and a link‑funnel diagram for monitoring. The resulting data outputs are assigned confidence grades to ensure reliability.

On July 10, from 14:00 to 17:30, OceanBase together with DataFun will host an AI “collision” event at Shanghai Ant S Space, where Ctrip experts will share the detailed implementation. Interested participants can register for free via the QR code provided.