Projects
Here are some of my projects. To view more details and others, please visit github.com/szujuchen.
Information Retrieval and Extraction Final Project | Dec. 2024
- Designed retrieval algorithms for legal question answering using related Taiwanese laws.
- Developed prompts and utilized LLMs to rewrite user queries and legal contexts to align their style for more effective retrieval.
- Trained a Bi-Encoder model and fine-tuned Taiwan-LLaMA-3B using original user queries and LLM-rewritten queries.
- Conducted experiments with various model settings, including a Bi-Encoder, fine-tuned LLM, and Retrieval-Augmented Generation (RAG).
- Accepted by the SIGIR 2025 Short paper
- Technologies: Fine-tuning LLMs, Bi-Encoder models, prompt engineering, RAG
Github Report Paper Poster
Computer Vision Practice with Deep Learning Final Project | Dec. 2024
- Developed a pipeline that combines visual grounding and image inpainting for seamless indoor object replacement.
- Fine-tuned models on a custom dataset for indoor segmentation and integrated them with state-of-the-art approaches like PowerPaint and BrushNet.
- Created a user-friendly interface and proposed solutions for successful object replacement and scene optimization using text prompts.
- Technologies: computer vision, fine-tuning segmentation models
Github Report Slides
Creativity and Entrepreneurship Program — Konnect | Sep. 2023 - Jun. 2024
- Designed a fan-connecting app to provide a platform for fans to interact with creators, post activities, and record memories.
- Led the full development of a cross-platform app, implementing features for activity tracking and content sharing.
- Built the app from scratch using React Native, integrating Google Maps, user posting, real-time data access, and a Google Cloud Platform backend.
- Delivered Minimum Viable Product (MVP) and passed initial testing
- Technologies: React Native, Firebase API, Google Cloud Platform (GCP)
Github Slide
2024 MakeNTU: Explorer Hackathon - Ctrl-F Autonomous storage robot | Mar. 2024 – May 2024
- Designed the working mechanisms of the storage robot and the overall user workflow within the mobile app.
- Managed app development and implemented the backend communication with the Raspberry Pi robot controllers.
- Supported the object recognition fine-tuning task and integrated the model into the app.
- Technologies: React Native, Firebase API, Raspberry Pi (RPI), computer vision.
Github Slide
