Hello Group (哈啰集团)

Solution

Hello Group built a transaction robot on Alibaba Cloud Bailian using a portfolio of Qwen-Plus, Qwen-Turbo, and Qwen2-7B-Instruct models. The robot spans pre-sales, in-sales, and after-sales steps of the mobility transaction chain, combining user-intent recognition, multi-turn conversation, intelligent Q&A, and liability handling into one business-facing service. To make the system practical inside production workflows, the team introduced RAG, chain-of-thought, and tool-calling patterns so the agent can retrieve policy or scenario context, reason across transaction states, and take guided actions in support and sales flows. Instead of treating the LLM as a sidecar chatbot, Hello embedded the agent directly into the transaction process, using different Qwen models for responsiveness and capability tradeoffs. That design turned the robot into both a service-efficiency layer and a revenue-oriented guide for rental scenarios.

Data flow

User requests from pre-sales, order-processing, and after-sales scenarios are routed into the Bailian agent layer, where the appropriate Qwen model is selected. RAG and tool-calling components retrieve policy or order context, the agent reasons over the transaction state, and the output is returned as guidance, liability decisions, or customer-service responses embedded in the business workflow.