Shenli Technology (深励科技)

Solution

Shenli Technology redesigned its AI companion stack around a layered Qwen architecture delivered through Alibaba Cloud Bailian. Instead of sending every request to one model, the company routes routine high-frequency interactions to Qwen3-flash for low latency and concurrency, while longer story generation, emotional counseling, or more difficult reasoning tasks are escalated to Qwen3-plus and Qwen3-max. Those model outputs are then passed through Shenli's product-specific prompt engineering, memory strategy, and action-driving logic so the same base model family can power two different companion products: AI Xiyangyang for children and AI Kitty Walulu for young adults. The model layer is embedded across input understanding, scenario recognition, response orchestration, memory accumulation, action tags for the physical toy, and automated content generation, creating a closed loop that balances latency, cost, character consistency, and creative freshness.

Data flow

User voice or text inputs are routed into Bailian, where Shenli's scene-recognition logic decides whether the request should be handled by Qwen3-flash, Qwen3-plus, or Qwen3-max. The resulting outputs are then transformed by role prompts, memory policies, and action tags before they are returned to the physical or app-based companion experience.