Infra DC SDE

1. Converted the Python code for garment defect detection to C++, and tried different ways to improve algorithm performance by more than 10 times

2. Designed defect detection backend service using Python Flask, interacted with frontend by GraphQL

3. Integrated different modules such as DL algorithms, front end, and back end together and deployed them on the local machine, ran a simple garment defect demo on Mac and iPhone as a webcam

4. Attended Advanced Textiles Expo in Orlando to promote the use of AI in advanced manufacturing to textile companies and fabric suppliers around the world

5. Expanded Sony remote sdk for controlling ILX-IR1 professional camera, interacting with the backend via Socket.io to take photos and modify parameters

Non-China commercialization technology, core data team, responsible for DataPi system research and development

1. Designed key field protection module of data flow platform, enabled Adlog to realize outlier marking and processing at field level, maintained data quality and controlled risks in different aspects

2. Ensured the independence and compatibility between ETL and CRUX index-dimension interfaces, adjusted approval process to achieve core data manual approval and non-core data automatic approval

3. Improved unit test coverage and developed USD exchange rate field splicing function of real-time dimension service, investigated and optimized using JavaRPC generic calls, achieved dynamic configuration of request parameters and decoupled from server side

4. Converged downstream links’ metric information, merged and integrated SQL automatic generation function in ETL fusion project of advertising data, and solved problem of multi-region one-click release, which greatly improved production efficiency and stability

Design AIoT(Artificial Intelligence of Things) application solutions based on different types of Espressif SoC boards(ESP32-S2, ESP32-S3), such as voice interactive robot dogs

1. Constructed a system that can output corresponding data websites according to the input geological keywords

2. Screened web pages roughly based on URL keyword matching, then used logistic regression and TextCNN model with metric learning to further classify

3. Extract meta information from web pages using the named entity recognition method, stored data into mongoDB