1. Built search engine of e-commerce website for Michaels Store

2. Developed backend API for search with Java and Elasticsearch

3. Developed semantic search service by encoding search keywords with sentence transformer and ranking the products based on similarity of embeddings

4. Improved semantic search performance by utilizing keyword extraction

1. Utilized embedding methods, such as GloVe, ELMo, BERT, FastText and Flair Embedding, and combined with sequence tagging model to do Named Entity Recognition (NER)

2. Achieved 92.95% F1 score for the best method on CoNLL 2003 English dataset

1. Designed a self-learning framework to segment objects in unseen video using Python and Caffe

2. Achieved favorable performance against the state-of-the-art algorithms (Our method: average IoU 72.6% on DAVIS 2016 dataset; TransferNet: average IoU 62.4% on the same dataset)

3. Published paper as the third author: “Unseen Object Segmentation in Videos via Transferable Representations”, Asian Conference on Computer Vision (ACCV), 2018