Ngo Jun Hao Jason

• Elective Focus: Artificial Intelligence
• Minor: Psychology
• GPA: 4.46 / 5.00

• Specialisation: TBD
• GPA: 4.00 / 4.00

An Augmented Virtuality Approach To 3D Videoconferencing

• Created a proof-of-concept for a 3D meeting app in Unity by applying concepts of narrowcasting

Deep Learning Based Car License Plate Recognition

• Fine-tuned a license plate detector (LFFD) with MXNet to get an average precision of 96.9% at an IOU threshold of 0.7
• Improved upon a license plate recognizer (LPRNet) using Tensorflow to reach an accuracy of 97.2%
• Combined the license plate detector and recognizer, resulting in a lightweight and fast license plate recognition system with an accuracy of 96.1%

• Built a website with Django that displays bounding boxes to aid the computer vision team in qualitative analysis of object detection models
• Deployed aforementioned website on an on-premise Nginx server
• Sped up the collection and pre-processing of more than 10000 images using Python scripts

• Contributed to the development of a website ( autoscrape.com ) built in Django that aids in lead generation
• Developed a prototype mobile app that controls iris doors remotely using Flutter and AWS (EC2 and Lambda)

• Created and enhanced APIs of several Spring Boot microservices for a customer relationship management website (Client Connect) that supplements the interactions between relationship managers and customers
• Used React to develop the UI of a financial modelling website (CLOVERS) as a proof-of-concept, intended to replace weeks of manual input in Excel spreadsheets with a few minutes of form submission
• Worked with site reliability engineers to set up CI/CD pipelines in Jenkins for multiple projects, replacing time-consuming manual checks with automatic stages
• Instilled a culture of documentation, reducing time spent on knowledge transfer and key person risk
• Onboarded several newcomers, helping them to get familiar with the code base and coding standards

• Developed an admin console with NextJS, FastAPI and Postgres that allows grant researchers to easily add business grants to the grant directory
• Integrated Voyage AI's multimodal embeddings and reranker with Postgres' pgvector to build a grant matching engine that finds the most relevant grants by ingesting project documents
• Implemented Retrieval Augmented Generation in a low-latency, low-cost fashion by using OpenAI's gpt-4.1-mini to generate explanations of why certain grants are good matches for a given project
• Created simple CI/CD pipelines using GitHub Actions, AWS Amplify and AWS App Runner to reduce toil and human errors in testing and deployment

• Digitised compendium for guests into a QR-accessible static website ( haastrivermotels.github.io/compendium ) that incurs no maintenance or hosting costs using HTML, CSS and JavaScript

• Used Zapier to automate the generation of custom manifests in Google Sheets from FareHarbor bookings as efficiently as possible, reducing up to 10 minutes of paperwork per booking

• Rebuilt the website ( kakapolodge.co.nz ) with more features using NextJS
• Saved around $3000 in unnecessary subscription fees annually via new websites and tech advice

• Rebuilt defunct website ( hanmerwoolshed.co.nz ) using Shopify, resulting in first online sales that made up 20% of revenue and expanding the business' reach beyond domestic market

• Developed landing page ( hanmerfoodway.co.nz ) with NextJS
• Generated inventory listings with the aid of Python scripts to avoid hours of manual effort

• Built simple and clean website ( somtam.nz ) with low ongoing costs using HTML, CSS and JavaScript