𝗔𝗜 𝗳𝗼𝗿 𝗛𝗲𝗮𝗹𝘁𝗵𝗰𝗮𝗿𝗲• Developed a transferability metric to predict AI model's performance in ECG diagnosis without actual training.• Developed a digitization software that turned paper-scanned images of ECGs into digital time-series data.• Built CI/CD pipelines on Azure DevOps for automating the training and deployment processes.

𝗠𝗮𝗿𝗸𝗲𝘁𝗶𝗻𝗴 𝗦𝗰𝗶𝗲𝗻𝗰𝗲• Conducted statistical analysis on survey data to generate key performance indicators (KPIs) and metrics related to client's brand awareness.• Automated quarterly reports using Google Cloud APIs.• Refactored and optimized SQL queries, improved query execution time by 22%.

- Developed Solidity smart contracts for NFT Game characters & actions, token sales, and staking rewards.- Tested & deployed the contracts on Binance Smart Chain.

Deployed CenterNet on Android for face detection using TF Lite C++:- Reduced model size from 7.18MB to 2.3MB by quantization- Used Java Native Interface (JNI) to integrate core C++ functions into the app

I participated in the AI Training Program. During this period, I developed a GAN for Image Super Resolution and a Transformer model for English-Vietnamese Machine Translation. GitHub codes are available.

Served as a Teaching Assistant for a Data Science course as part of the Developer Circles - Vietnam Innovation Challenge, which was organized by Facebook and CoderSchool.Assisted lecturers with in-class activities, graded coding assignments, and provided personalized support to students outside of class time.

Built and trained advanced neural networks to classify complex multi-label datasets in 12-lead ECG recordings for the diagnosis of multiple abnormalities. Utilized cutting-edge techniques, including CNNs, RNNs, and Transformer Encoder, achieving 0.96 accuracy on precision-recall curve with the combination of Residual Convolutions and Transformer Encoder.

𝗣𝗿𝗼𝗷𝗲𝗰𝘁𝘀 𝗰𝗼𝗻𝗱𝘂𝗰𝘁𝗲𝗱:𝟭. 𝗥𝗲𝗮𝗹 𝗘𝘀𝘁𝗮𝘁𝗲 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁• Built, trained, and tuned convolutional neural networks for image classification. Improved $f_1$-score to 0.9.• Deployed trained models to AWS SageMaker Endpoints.• Built wrapper APIs with AWS API Gateway, Lambda Function and DynamoDB.• Developed back-end web application with Django and PostgreSQL.𝟮. 𝗢𝗻𝗹𝗶𝗻𝗲 𝗿𝗲𝗰𝗼𝗺𝗺𝗲𝗻𝗱𝗲𝗿 𝘀𝘆𝘀𝘁𝗲𝗺• Implemented collaborative models (Alternating Least Square, Logistic Matrix Factorization). Improved click-through-rate by 11%.• Utilized AWS SageMaker for hyperparameter tuning and model deployment.• Developed and dockerized backend APIs for production on AWS Fargate.

Project: 𝐌𝐚𝐜𝐡𝐢𝐧𝐞 𝐋𝐞𝐚𝐫𝐧𝐢𝐧𝐠 𝐟𝐨𝐫 𝐏𝐥𝐚𝐬𝐦𝐨𝐧𝐢𝐜 𝐝𝐞𝐬𝐢𝐠𝐧• Objectives: Finding relations between the nanostructure of graphene-based metamaterials and their optical spectra, then solving the inverse problem: designing materials which satisfy given spectra.• My roles & contributions: Applying regression techniques (Support Vector Regression (SVR), Deep Neural Networks)to solve the inverse problem: given the optical spectra, find the structure of the materials.

• Developed an infection screening system for biomedicine based on Machine Learning and signal processing algorithms to filter out noise in multiple vital signs extracted from a medical radar and detect signs of infection.• Designed and implemented an IoT-based smart water meter system, leveraging number-recognition algorithms running on STM32 microcontrollers to detect water meter numbers. Utilized LORA protocol to transmit data to a server for further processing.• Achieved exceptional results in Vietnamese Sentiment Classification, leveraging advanced Bidirectional LSTM techniques to classify a large dataset of customer sentiments into 3 classes: positive, neutral and negative. Delivered f1-scores of 94%, 87%, and 91% respectively.