Projects
Selected academic and applied AI projects across multimodal document intelligence, handwritten mathematical expression recognition, educational AI, computer vision, data systems, and model deployment.
Research projects
Multimodal Handwritten Mathematical Expression Recognition
Developed multimodal deep learning methods for handwritten mathematical expression recognition (HME recognition), combining online stroke sequences, offline rendered images, and structural relation modeling. The work includes encoder-decoder models, graph-based relation prediction, weakly supervised offline recognition settings, and symbol-structure analysis for robust mathematical expression understanding.
Domain-Specific Language Modeling for Handwritten Mathematical Expression Recognition
Investigated mathematical language models for HME recognition, including n-gram and Transformer-based models for improving structural and syntactic consistency during recognition and decoding. This work connects visual recognition with domain-specific sequence modeling for mathematical notation.
Edge-aware GAT for Handwritten Mathematical Expression Structure Recognition
A graph neural network model for predicting spatial relations between handwritten mathematical symbols using edge-aware attention and relation decoding for structure-aware handwritten mathematical expression recognition.
Automatic Scoring for Handwritten Descriptive Answers
Research on automatic scoring systems that combine handwriting recognition outputs, answer similarity, confidence-based rejection, and human-in-the-loop verification, with ongoing exploration of LLM-assisted semantic scoring for long-form answers that require meaning-level understanding.
E-compass Automatic Scoring for Geometric Construction
A process-aware scoring framework for geometric construction answers using digital ink from an electronic drawing compass, stroke-level clustering, step-level segmentation, and geometric reasoning.
HAIMR: Handwritten Answer Input, Management, and Recognition for Moodle
A Moodle plugin that supports handwritten answer input and recognition for Japanese, English, mathematical expressions, and chemical formulas.
Industry and applied AI projects
Social Media Data Platform and Analytics
At HIIP Asia, I worked on a company-wide platform that collected social media data and supported downstream analytics for viral marketing and campaign insights.
Brand–Influencer Recommendation System
Recommendation models for matching brands with suitable influencers based on social media data, user profiles, content analysis, and campaign requirements.
Image and Text Classification for Social Media Data
AI models for analyzing crawled social media content, including image and text classification for extracting useful signals from large-scale online data.
Sushi Recognition
A computer vision model for detecting and recognizing multiple types of sushi in real-world images.
Japanese Wine Classification
A fine-grained image classification model for distinguishing more than 200 types of Japanese wine.
Asian Face Recognition
A face recognition model focusing on Asian face datasets and real-world recognition scenarios.
Systems and deployment
Handwritten Mathematical Expression Recognition Server
An applied AI server for handwritten mathematical expression recognition and handwriting recognition for mathematics, English, and Japanese, supporting online and offline inputs through deployable APIs.
Prototype AI-Agent Workflows
Explored small-scale AI-agent demo workflows for research assistance, coding support, and task decomposition. These prototypes were developed for personal experimentation and workflow exploration rather than production deployment.
Traffic Video Analytics
Systems for estimating and visualizing traffic parameters from video, including vehicle detection, tracking, and real-time video processing.