CV

Download the CV via this link

Education

CocoRobo Ltd.

2023. 06 - 2023.08

Built up a customized chat-bot for online education platform used by 20+ schools using Langchain and accomplished a pipeline from indetifying question to responding question.
Significantly improved response-time of the chat-bot by 30% by applying Langchain Agent and integrating GPT-3.5 API.
Enhanced the robustness of the chat-bot by 50% by doubled the variety of files that could be processed by introducing the vector database and prompt engineering.

TsingHua Berkeley Shenzhen Institute

2021. 12 - 2022.03

Accomplished a baseline of correction and reconstruction of MRI 3D image data.
Tested a new MRI prepossessing baseline by reducing the segmentation error for around 5%.
Deploy U-Net architecture for MRI data segmentation and reconstruction via Pytorch. Successfully run the whole pipeline and received available result in real-world application.

Master Project at University of Zurich

2024. 07 - Present

Develop and deploy AI-based function on the digital education assistance platform (KlickerUZH) of the university. The agent will generate questions based on the lecture materials provided by lecturer.
The answer will be generated by GPT-4o model. We will do prompt engineering to let it adapt to multiple functions such as question types, question difficulty and multimedia.
item We will develop a evaluation pipeline to test the quality of the question generated and used for creating own dataset used for later potential fine-tunning process.

semester project at ETH Zurich

2024. 02 - 2024.07

We adapt existing Visual-Language Model (VLM) methodologies to analyze city-scale scenes, focusing on detecting urban characteristics.
The building segment task attains an AUC score of 0.927 in specific scene, which is better than existing pipeline (0.875 AUC).
Our framework indicates that VLMs have the potential to solve urban classification and localization tasks, especially good understanding of some abstract phenomena.

course project at University of Zurich

2023. 10 – 2023. 12

Build up a full-stack chat-bot from scratch to answer questions provided by human user and offer recom- mendation given user’s preference from Wikidata’s movie database.
Find the answers and generate human-like response by a mixed mode with SPARQL query, crowd-source research and BERT model. Handle irregular user input by Spacy during entity and relation extraction. Deploying the chat-bot to website by Speakeasy API.

Programming:
- Python; C++; R; SQL; Linux; Git
Framework:
- Pytorch; OpenMMLab; Scikit-learn; Langchain; Spacy; NLTK; MySQL
Office:
- Microsoft Office (World, Excel, PowerPoint); LaTeX; Markdown
Language:
- Chinese (Native); English (Work Proficiency - IELTS 7.0); German (Elementary)