- Research and development of personalized restaurant recommendation system using GNN-based recommendation and LLM-based text analysis.
- Develop a more reliable ranking algorithm by incorporating Bayesian approaches and other statistical weighting methods to reduce noise from extreme or manipulated reviews.
- Image Preprocessing Pipeline – Built a structured pipeline to quantify various facial attributes (face-width ratio, orientation, eye/mouth geometry, hair position) and select the highest-quality images.
- Stable Diffusion & Face Swap Integration – Incorporated model-based generation for refined user likeness, applying Gaussian blur where appropriate to ensure a natural skin appearance and overall consistency.
- Facial Geometry & Shape Analysis – Implemented robust algorithms (via Google MediaPipe FaceMesh) to detect 468 landmarks, enabling high-precision face shape classification.
- Personalized Makeup Guidance – Developed per-region analysis (contouring, highlighting, etc.) and offered example styles based on personal color theory, utilizing a flexible data flux approach for adaptive content.
- Development of open software to support victims of digital sex crimes.
- Bronze Award at the 2022 SW Developer Competition.
