• Led the development of the Labeling Task Generation Self-Server UI via React and Flask to streamline the process of

requesting, generating and organizing data annotation platform

• Launched permission control and monitoring system via OneLogin Single Sign-On

• Organized and queried info from MySQL database, based on front-end user input

• Improved success rate and stability of Task Generation Pipeline including data retrieval, transformation, processing and uploading

• Greatly accelerated the development speed by building up local test environment for Task Generation Pipeline

• Modularized Task Generation Pipeline code to improve code extensibility

Optimization of thermal power generation efficiency

• Summarized data from 192 types of sensors and conducted pre-processing process on around 12 million samples on this base

• Designed Reinforcement Learning Network based on Actor Critic to predict future states according to 94 dimensions’ actions

and 98 dimensions’ states data

• Built Simulator Model to save past data and generate new data, in order to enlarge the training dataset

• Deployed a Loss Network in the model to set up specific limitations during boiler operation

• Analyzed operation situation of steel ball lapping machine according to 520 thousand 77 dimensions’ samples via LSTM

AI Platform

• Integrated and modularized code related to water quality monitoring model based on LSTM method

• Designed CNN-Based Unsupervised network to do painting clustering via Torch Network Structure in Lua Language

• Built AI-enhanced art recommendation engine via Supervised Convolutional neural network

• Optimized data base structure and added tags to paintings