Experience

Ycomm Việt Nam

• Vietnam
• Software Development
• 1 - 100 Employee

• Machine Learning Engineer

  • Apr 2023 - Present



VinBigdata

• Vietnam
• IT Services and IT Consulting
• 100 - 200 Employee

• AI Engineer

  • Sep 2021 - Jul 2022

  • Second Prize – NLP Hackathon at Biomedical Center of VinBDI. • Visual Question Answering: I adopted bottom-up-attention network, whose feature extractor is Faster R-CNN object detector. Each word in the question is transformed into a vector using the 300-D GloVe word embeddings. The word embeddings are then passed through a one-layer LSTM network. I also replaced word embedding with BERT. • Vietnamese Spelling Error Correction System: I used word augmentation to generate the Vietnamese spelling error data which is the target data training. I labeled correct and mistaken tokens by BartPho tokenizer. For the model, I researched word generation that I installed Sequence to Sequence (RNN, Bi-LSTM, GRU) combined attention and the Transformer. Moreover, I researched and implemented the Soft-masked BERT. • Text to speech with subword embedding: I researched combining phoneme context with subword context using Tacotron2 model. I programmed the stepwise monotonic attention for extracting context. I implemented a tokenizer for Vietnamese using Huggingface tokenization and Vietnamese data and used that tokenizer to generate subword embedding which concatenated with BERT CLS token replication. I also used K-means clustering and cosine similarity to analyze the context of BERT CLS token and then I could synthesize the speech into 3 different styles. Show less



Fpt Telecom

• Vietnam
• Telecommunications
• 400 - 500 Employee

• Data Scientist

  • Jun 2021 - Oct 2021

  I researched the Unknown Recognition by camera project. I researched replacing the backbone of Retina face with Efficient Net. I researched the ArcFace-Additive Angular Margin Loss for feature extraction. I installed Faiss to optimize the register and the adaptive threshold for label classification. I installed human detection and Person Re-ID to enhance the performance. I researched the Unknown Recognition by camera project. I researched replacing the backbone of Retina face with Efficient Net. I researched the ArcFace-Additive Angular Margin Loss for feature extraction. I installed Faiss to optimize the register and the adaptive threshold for label classification. I installed human detection and Person Re-ID to enhance the performance.



Cloud Nine Solutions Company

• Vietnam
• Software Development
• 1 - 100 Employee

• AI Intern

  • Jul 2020 - Oct 2020

  I collaborated with another intern partner to implement the Facial Recognition Check-in. I used Dlib to detect faces in image, then I chose one closet face. After that, I aligned that face before extracting by 2 dense layers and I classified label by SoftMax. I built a video streaming by Flask to demonstrate the project. I collaborated with another intern partner to implement the Facial Recognition Check-in. I used Dlib to detect faces in image, then I chose one closet face. After that, I aligned that face before extracting by 2 dense layers and I classified label by SoftMax. I built a video streaming by Flask to demonstrate the project.