In this article, we design a robo-advisor which has a bi-level framework. The framework enables it to handle a large amount of assets using fast algorithms in the lower level. The proposed robo-advisor can utilize the closed-form solutions for investors’ risk preferences based on corresponding portfolio choices. A dynamic weight is applied to update investors’ risk preferences. Numerical results based on real data in Chinese stock market show that our proposed robo-advisor can accurately estimate the risk preferences of investors and outperform the benchmark formed by market indexes.