aiTimelog• Led the design and development of aiTimelog, an AI-driven attendancemanagement system, comprising the AI Box and Web Server components.• Leveraged face detection, facial feature extraction, and vector similarity search to accurately detect and recognize employees from camera feeds. Provided real-time recognition results on a customer-side app and transmitted data to the Web Server for processing.• Designed a scalable back-end system using Django Rest Framework andNextJS to handle customer data (Employee, Shift, Schedule, etc.) and real-time AI results from the AI Box. Calculated employee attendance and generated detailed reports, significantly enhancing HR productivity for clients.• Tiles Quality Control for CMC Tiles• Collaborated with the customer to identify key pain points in the tile production process.• Led a team to develop an AI-powered system using Edge Computing on Jetson Nano, integrated with the customer’s PLC system to detect defects andmismatched tiles, reducing verification time andmanpower by 90%.

Attendance Management System• Led a team to develop a real-time face recognition system for AI cameras, utilizing person detection, tracking, and face recognition to enhance attendance tracking accuracy.• Researched and developed a facial feature extractionmodel that has the ability to accurately extract facial information from face covered and not covered bymask - a 800% increase in accuracy in recognizing masked face.• Designed an end-to-end system incorporating batch computing and parallel processing, enabling multi-stream processing in real-time and achieving a 300% increase in FPS.Online Media Face Search• Built clustered searchmethod and distributed data system for face searching system using social media photos and video on Youtube as input. System required to process 3000 videos a day.• Designed and developed an end-to-end system that detect face in videos and photos, extract feature embedding, and query search in a database contain 10million different facial vectors.Crowd Behavior Recognition• Developed amodel that detect people accurately on slightly crowded scene and group people using clustering algorithm based on distance - achieved high accuracy (test on school datasets)• Created amodel to classify violent and non-violent actions in crowd settings - achieved high accuracy (test on school datasets) and enhancing customer security measures

Viettel AI eKYC• Developed amodel that detect faces, extract facial information and compare faces of customer ID card, Passport, Driver License with real life facial images using the extracted facial features - achieved a near perfect accuracy in matching faces.• Created an image verification process for eKYC systems to detect anomalies like sunglasses, blurriness, closed eyes, and improper lighting, enhancing the reliability of input face images.

Government Identification Paper OCR• Developed an eKYC system that extracts information from Vietnamese Identification Documents such as National Identification Cards, Passports, and Driver’s Licenses, reducing manual input and verification time by 70% for customers• Expanded eKYC system capabilities toMalaysian and Indonesian IDs, enabling market entry into other Southeast Asian countries.