Report to ProfessorProject: Heuristic Semantic Search.• Description: Users can query the most similar patterns by meaning.• Technologies: Python, NLP, K-mean, Sentence Transformer, Sentence Vector, Heuristic Tree, NLTK, Deep Specific Search, Cosine Similarity.• Responsibilities: process pattern documents, data structure and AI model.- K-mean to cluster patterns in a class to extract the central meanings, then enrich the heuristic information field of this class node.- Search algorithm can stop at the most suitable node, then search all patterns in class to find out the most similar patterns by cosine similarity.• Achievement: Proposed heuristic search tree to search faster and more accuracy. Three journal papers are under review.

Report to ProfessorProject: Credit Scoring (4 members: 1 reporter, 1 data processer, 1 programmer, 1 AI developer)• Description: Write functions to score the credit of users based on rules from bank, classify customers into groups, KNN to test classifying result, Deep Learning to predict credit score.• Technologies: Python, K-Nearest Neighbor, Deep Learning, Ubuntu• Responsibilities: handle the AI models.- Build KNN model to re-test the result between rule-based classifying and clustering.- Build ANN model to predict the credit score.• Achievement: KNN accuracy 97%, CNN predict score loss 0.14Link: https://medium.com/@duydatspk/credit-scoring-using-ai-9548881209c7

Report to Vice President & Tech LeadProject: Develop a real-time health-care AI system, it can diagnose heart and lung diseases by sound (9 members: 1 product manager, 1 technical leader, 1 hardware coder, 1 hardware designer, 1 UI designer, 1 competitor analysis, 1 mobile developer, 1 network security, 1 AI Engineer)• Description: Build up AI health-care system from mobile app to speaker (hardware + software), client-sever connection, sever handles preprocess audio & AI model to classify the heartbeat parts to suitable diseases.• Technologies: python, audio libraries- Audio processing: librosa, numpy, pandas, scipy- AI models: CNN, ResNet, Tensorflow- Algorithms: Combined Moving Average, Fast Fourier Transform, Auto Cut-Off Frequency, Low-Pass Filter, Logarithm Spectrogram, Data Scaling by value and time, Heartbeat Segmentation based on systole distance.• Responsibilities:- Follow and Enhance the previous procedure from audio processing to AI model by more efficient method.- Research highly cited Q1 journal papers to find optimal solutions.- Tracking wrong-label samples via predictions from two different DL models, then give them the most suitable label.- Discuss & Report progress to team leader & vice president of corporation.• Achievement: - Val_acc improved from 80% up to 99% for general class, 60% up to 85% for specific diseases in a class of similar diseases.- Val_loss reduced from 0.441 to 0.025.- Number of epochs reduce from 600 to 25.

Report to Tech Lead, CEOProject: Explore Customer Behavior via LRFMP (1 Data Analyst)• Description: Detect customer behavior, predict items and sales, propose suitable solutions to the sales team.• Technologies: back-end python- Preprocess data: scikit-learn, scipy, matplotlib, pandas, numpy- AI model: LSTM for revenue prediction- Algorithm: LRFMP – Length, Recency, Frequency, Monetary, Periodicity• Responsibilities: - Clean data and find out shopping patterns of retailers, classify users.- Predict revenue, best-sold group of items in near future.• Achievement: - Boost up sales via discussing with sales department about phenomenas on data.- Find out diverse buying patterns in a week between VIP retailers and small retailers.- Three groups of customers: Best buyers, VIP potential, low frequent shoppers.- Predict revenue and goods in near future with low loss, less than 10%.

Teaching students about Graph Theory, Database, Discrete Mathematics, Dynamic Website Programming.

Report to Head of DepartmentProject: Hydraulic Computation Software at Mechanical Fluids Department (6 members: 1 professor, 1 team leader, 4 students)• Description: Develop UI, handle functions for core library, animation for the hydraulic software.• Technologies: C#, Window• Responsibilities:- Get the requirements from professor and create software architecture.- Design User Interface, work schedule, data flow in core library.- Guide students in team to code half of specific functions for different hydraulic structures.- Code half of processing functions for common hydraulic structures.• Achievement: Project has been passed by school scientific committee.