In the summer of 2019, I interned at Cloud Xiaowei AI platform for Tencent in Shenzhen, China. Cloud Xiaowei is a department composed of scientists, engineers, designers and product managers developing products, services, and experiences based on speech AI technology, such as smart speakers.

During my time at Cloud Xiaowei, I worked as a AI product manager. My work focused on  improving user experience of Tencent Dingdang through interaction design, a smart speaker similar to Amazon echo. During my two-and-a-half-month internship, I was mainly involved in two sub-projects on Tencent Dingdang.

Improving UX on Question Answering about Persons


structure of a typical dialogue system

Users ask all kinds of questions to smart speakers in their daily life, varying from information retrieval, such as weather forecast, to chatting such as requests for a joke. Our smart speaker, Tencent Dingdong is supported by a large dialogue management system inserted with NLP algorithms, under which various these queries are distributed to different skills ( smart speaker’s ability to do a kind of job ) .

During my internship, I was responsible for improving users’ satisfaction on smart speakers’ answers on famous persons. Data on bad queries ( user queries which don’t receive useful answers ) were analyzed and categoried. Then several new intentions ( ability to answer a specific kind of questions ) for smart speakers were designed to improve user experience. The initial ideas were presented in the form of tables and sketches. They briefly showed user experience under different circumstances. I then assessed the  feasibilty and value of further technological development by making a trade-off between investment and earnings.

a debug interface for AI product managers

Finally, several NLP algorithms for better understanding of questions on famous people were purposefully designed after rounds of discussion. Now they are already available on the latest version of Tencent Dingdong smart speaker.

Multimodality Speech Assistant for Children

In China, a growing number of parents regard smart speakers as a home tutor for children to learn knowledge. Smart speaker usually interacts with users through audio output. However, visual information is much more attractive to children than adults, the absence of which distracting children from interacting with smart speakers for a long time. Thus, our team explored some multimodality ways of interaction especially designed for children.

A 3D assistant image was created to accompany young children. We also created storyboards where children interacted with the image under a number of scenarios, such as storytelling and calling the bed. New interaction processes were also designed to apply full duplex (a new technology which free users from wake-up words such as ‘hi, siri’ ) into the new image. The project was still under way when I left Tencent.


Being a product planner in a large company is very different from scientific researches at universities. What really impressed me a lot is the number of discussions it took to reach to a small conclusion, sicne every action must be provable and accurate. One need to consider every aspects from market to cost, user experience, difficulty of tech realization,etc..

The internship experience not only equips me with basic knowledge of product management, but also provides me with new lines of thought on interaction design.