Willard Khoa - AI Engineer Team Lead

Highly motivated and experienced AI Engineer with 7 years of experience in artificial intelligence and a deep commitment to advancing the field. Skilled in GenAI, Machine Learning, Deep Learning, and Natural Language Processing with a proven track record of delivering high-quality solutions that drive business value. As a Kaggle Master with an impressive track record-1 gold, 8 silver, and 3 bronze solo medals. Exhibited outstanding expertise in tackling intricate problems and crafting cutting-edge AI solutions Proficient in a range of programming languages, including Python, Java, and C++, with expertise in popular Al frameworks such as TensorFlow, and PyTorch. Strong understanding of data structures, algorithms, and software design patterns, with excellent problem-solving skills and attention to detail. Collaborative team player with excellent communication skills, able to work effectively with cross-functional teams to drive project success.

Project list

Auto Review and Auto Score for Stock Photo & Footage

In a stock platform, the quality control of each item is very important. This project help reduce about 50% work load of manually reviewing effort of labeling team, by applying AI to automatically review and score on 100M items

Develop a scoring algorithm that accurately assesses performance based on the defined criteria Integrate the scoring system into the existing platform or application Develop techniques to identify content generated by AI models, such as analyzing language patterns, coherence, and consistency Auto Score: Person in charge, plan, develop, review Auto Review: Generative AI content detection, Abnormal event detection, Optical Flow analysis Implement a system to flag potentially generated content for further review Utilize machine learning algorithms to detect unusual or anomalous events within the data Monitor for patterns or deviations that may indicate fraudulent activity or system failures Employ optical flow techniques to analyze video data and identify changes or movements Detect abnormal behavior or objects that may indicate security threats or safety hazards

Search by Image

Instead of typing keywords, users sometimes want to search for images with similar content to their own images. This project can solve the problem by applying a lot of techniques from Image Retrieval domain. This feature is integrated into the stock platform with a database of 80 million images.

Oversee project planning, execution, and progress tracking, ensuring adherence to deadlines and budget constraints Lead technical design and solution architecture, proposing and implementing innovative technologies like CLIP and Faiss to deliver a high-performance and accurate image-based search system Develop and implement a deep learning-based image feature extraction model using architectures Integrate the image search system with the stock platform's database of 80 million images, using APIs or data pipelines Train and evaluate machine learning models for image feature extraction, indexing, and similarity search

Building Machine Learning Testing system

This project is a cooperation between AI Department and Test Department. The project aims at building a comprehensive solution to test AI systems in different use cases.

Management: control plan and progress, discuss with HR Team Adversarial attack and defend (working with VNU-UET) Test set generation: Augmentation, Diffusion, GAN Robot grasping testing MLOps testing system

Test Scoring for Entrance Exam

Aiming at diminishing the manual effort of HR Team when scoring interviewees' entrance test in the recruitment process, this project provides an internal tool to automatically evaluate test scores for every candidates.

Management: control plan and progress, discuss with HR Team Template detection Checkbox detection Digit handwriting recognition Review code quality

Electrical Drawing Reader

This is a project from a Japanese Electrical Corporation. The solution can automatically convert raw electrical sketches into drawing files of an existing Electrical Software. It can reduce a lot of work load for Electrical Engineers.

Management: control plan and progress, discuss with customers Data labeling and generation: apply pseudo-labeling methods to diminish labeling effort, generate synthetic datasets for training. Symbol Detection and Recognition Review code quality

Label OCR in POS machine

A demand from Japanese supermakets: integrating OCR into the POS machine to automatically applying discount labels for customers' commodities.

Design algorithms to filter low-confident prediction Design augmentation scheme to improve Text recognition Review code quality

Product Recognition project

This project is a cooperation between AI Department and Test Department. The project aims at building a comprehensive solution to test AI systems in different use cases.

Management: control plan and progress, discuss with customers Data filtering and generation: synthetic data generation, Image Quality Assessment Product detection: YOLO v7, DINO Product recognition: EfficientNet, Swin Transformer, arcface Review code quality

Embedded AI Development

The purpose of the project is building lightweight object detection and classification models to solve autonomous driving problem. Trained models are then quantized and integrated into MediaTek i500 board.

Object Detection and Recognition with Lightweight models: EfficientDet Lite, MobileNet, SSD, YOLO, etc. Quantization techniques: float32 to int8

akaDoc OCR Platform

akaDoc is a comprehensive document digitization solution. It is applied in a lot of business, such as banking, insurance, etc. to automatically extract and store essential information from diverse type of documents.

Document Detection, Retrieval, and Classification Script Detection: Latin, Japanese, Chinese, etc. Text Recognition: Latin (Lite version), Japanese. Optical Mark Recognition (Checkbox Detection)

Project list

Education

Techstacks