Not safe for work model for live streaming platform GGLive* Crawl, process and analyze data in accordance with application standards* Train, debug model( using CAM to visualize class feature then update data) to improve accuracy* Deploy model to production: Thread, batch prediction• Build a content distribution system created by ChatGPT according to business requirements

• Researched on computer vision problems: object classification, object detection, object segmentation, optical character recognition and face recognition• Build and deploy censorship model (Image & Video) for the company’s social network and media (accuracy = 97%) * Models: Not safe for work, horror classification, reactionary classification,.. * Using deep learning model such as CNN, YOLO, Mask R-CNN * Design censorship architecture for Lotus app, internal newspapers * Technical: FastAPI, Kafka, MySQL, PyTorch• Created and deployed a system to recognize Driver License, Health Insurance (accuracy = 95%) * Deep learning model: YOLO, DB Net, Transformer OCR, LayoutLM * Converted model transformer OCR to TensortRT( reduced more than 2 times inference compared to PyTorch )• Build celebrity recognition API support over 895 profiles (accuracy 92%) * Using deep learning models: SCRFD, ArcFace, HNSW * API support: recognize, add new celebrity person•Face check in system * Recognize face employee from company’s camera * Build API for called from mobile app• Image captioning for Vietnamese * Crawl, processing data( COCO image caption 2017, Facebook ) * Models: Resnet50 + LSTM + Attention, Faster R-CNN + M2 transformer• Have experience using Kubeflow and BentoML to create ML workflow, manage and deploy models on K8S GPU cluster * Create ML pipeline: data processing, training using KubeFlow * Manage and deploy model using BentoML