Implemented model server deployment service on web console & resource uploading service for user-created clusters.

Worked on all front-end (JavaScript, React, Redux), back-end (Go, PostgreSQL) and CLI (Python) development.

Using python to collect and clean data from public datasets, as well as maintaining entry data for 1,000,000+ images

Designed neural networks specifically for scene classification tasks with Python. Implemented teacher-student

Devised a view discerning unit to adjust the weights of different views, which improved the 3D shape retrieval performance of the unweighted network by about 4% on Mean Average Precision (MAP).

Extracted spacial structural information from a set of image features through recurrent neural networks and got a performance boost of about 5% on MAP, compared to naive feature merging techniques.