Mainly focus on model optimization and model deployment for problems relate to AI applications for medicine. Leverage moderns techniques for model optimization (quantization, pruning,...), modern engines (TensorRT, OpenVINO, DeepStream, Holoscan, OnnxRuntime, TritonServer, ...) and CUDA programming to speed up model on production.

Build a text classification model for user queries to coarsely detect user intent. Achieve 96.6% F1-score.Build a query understanding model (named-entity recognition) + knowledge base graph.Build an auto correction model based on deep learning method and n-gram language model for correcting user queries. It achieved 45% performance w.r.t google auto correction for misspell cases.Research and implement Learning to Rank model (XGBoost, Semantic Product Ranking)Design a complicated dataset flow for the model (include airflow, bigquery, Hadoop, Spark, K8s).Design and implement AI infrastructure on AWS cloud for automatically aggregating data from multiple sources, training and evaluating machine learning models.

Leading Machine Learning EngineerResearch, design and implement a framework on Tensorflow (version 0.11, 0.12, 1.0, 1.1, 1.2) allow to write fastly a end-to-end machine learning models based on Tensorflow. It’s same as the architecture of `tf.Estimator` with respect to the later versions of Tensorflow.Implement some the state of art deep learning models in computer vision (Resnet, ResNext, DenseNet, Inception,...) for common problems as image classification, object detection, image segmentation, image similarity. Specifically, wrote Mask-RCNN with feature pyramid support from scratch on Tensorflow when no public code available.Research and implement a end-to-end NLU (Natural Language Understanding) model to convert a natural sentence to a lambda expression. Model pipeline include sub-models: text classification, NER tagging, auto-correction, semantic parsing. All codes wrote based on pytorch, scikit-learn, antlr4 python packages.Research, design and implement a task manager system is able to automatically scale, train almost machine learning tasks on kubernetes cluster and AWS EC2 cloud, it make all applied machine learning tasks in the company running through RESTFul APIs.Research reinforcement learning, recommendation systems and some generation model based on deep learning techniques

I'm self-learning at my home about AI, Machine Learning fields.

Felizz is a small startup company. I have worked at here as a key member. My work is the developing backend of web products of the company. Based on NodeJS ecosystem together with MySQL and MongoDB.

+ Development Data Leaking Prevent (DLP) system to prevent the leaking secret data, documents from a personal computer based on the hooking API technique, plug-ins on common software as Microsoft Office, Microsoft Outlook.+ Development Web-based platform for the Management Switch System, backend and frontend. Backend is wrote by Python and interact with Elastic-Search database. Frontend is wrote by the AngularJS framework

+ Analysis malware scripts+ Development the automatically analysis malware scripts system. Emulating Windows OS system - system APIs, memory management system, process management system+ Research Windows OS kernel