NLP
• Trained Google's Transformer T5 model on BillSum dataset to summarize an input paragraph.
• The model achieves equal score to Google’s original T5 paper.
• Built a Django-hosted web page for demonstration, SQLite to store summarization history.
• Extracted the Encoder Vector from the Text Summarization model to represent article content.
• Compared the cosine similarity of the vectors of multiple news articles to get the most relevant articles based on the one you pick.
• Adapted the system for Plagiarism detection based on textual similarity.
Computer Vision
• Intergrated Yolo model to backend server to detect object in image.
• Added Opencv Object Tracking to reduce computation needed for video processing.
• Reduced computation needed by 95% while maintaining accuracy.
• Combined with Depth Estimation model to compute a suitable path for collision avoidance.
• Combined Vision Transformer with Language Model to create a locally hosted image/video captioning server that works with both mobile phone app and desktop web page.
• Hosted the demonstration page on RESTful Flask server with PostgreSQL database.
• Integrated HuggingFace’s ViT model to the aforementioned Flask server.
• Ulitlizing React on the desktop webpage to communicate with the server, with user's registration and login scheme, user private directory, private links, captioning history…
• Used OpenCv on C++ to track a paper marked with Aruco Marker and normalize the position of content inside it, regardless of orientation and photo angle.
• Application in human pose tracking, QR code reading, courier service label reader.
• Adapted ResNet model to train on custom dataset to segment “person” class from a video.
• Combined with MiDaS depth estimation to create an estimate of depth in a video.
• Aiming to generate a side-by-side VR videos to stream on Oculus Quest.
• Trained a U-Net Deep Learning model from scratch on the COCO dataset.
• Enabled the model to segment classes of objects from each other in an image.
• Visualizing the generation of SuperPixel from image, showing the execution step-by-step of the SLIC algorithm.
• Optimized the algorithm's speed with Numpy’s Matrix Multiplication instead of a traditional for-loop, yielding a 50% speedup compared to the original codebase.
• Forked an original, non-working repository for a JavaScript project about utilizing OpenCv2 on JavaScript to “reverse-perspective-warp” a 2D plane in an image.
• Fixed bugs and logic errors to make it functional, and added a Flask demo for the algorithm.
• Experimented with Python and OpenCV to process live webcam images, capturing laser dot movement and converting it into mouse movement for a remote, touch-free mouse operation.
• Pre-processing input images to standardize it and reducing noises.
• Utilizing Principal Component Analysis to generate Eigenvectors from a dataset of face images, encoding each faces as a combination of Eigenfaces.
• Usable in Facial Recognition System, or lossy data compression down by 70% of the original file size.
• Converted manufacturer’s IO shield image to a 3D printable model file.
• Apply image morphology algorithms with OpenCv.
• Deployed the application with Flask and Docker on a self hosted machine.
General Project
• Developed a Coursera-style learning platform as part of a sophomore project.
• Built a Node.Js application with Agile software development methodology.
• Developed an app checking for the user's total playtime and shut down the game if it exceeded the user-defined threshold to combat League of Legends gaming addiction.
• Used PyQT for user interface.
• Gaming agent utilizing Alpha-Beta pruning search algorithm to efficiently play Fox and Geese board game.
• Optimized heuristic contains Potential Field Function, Breath-First Search, Exploration-Exploitation Balancing.
• Achives second place in class competition.
"Project"?
• Coded the Baillie–PSW pseudoprime test for the BigNum problem in C++.
• Coded BFS, DFS, A*, Heuristic search Algorithm in Python.
• Excel folder content fuzzy search.
• Competitive Programming experience.
• Built an Arduino CNC Shield to build a homemade cnc machine, 3D printed an adapter for the plasma torch and control the torch pulse using Arduino, wrap the controller board in E.M.I resisting material and successfully deployed the machine for real world application.
• Built a Web hosting/Nas machine running Debian Mint Linux.
• Hands on experience with Apache, Nginx, SSL Certification creation, multi-drive configuring, Docker.
• University coursework required me to master basic office suite like Canva, Latex, Markdown, Microsoft Word/Excel/Powerpoint.
• Used Sketchup quite a lot to do 3D modelling, made many practical models with it.