Jinghe Wang Daniel Wang
Computer Science and Engineering (Artificial Intelligence) undergraduate at HKUST, building production-oriented LLM workflows and native applications across iOS and visionOS. Fluent in English and Mandarin.
My work spans structured LLM data pipelines, multimodal AI integration, and 3D spatial computing on Apple platforms, with an active exploration of LLM agents.
- Education
- BEng in Artificial IntelligenceHKUST · Expected Jun 2029
- Distinction
- Full ScholarshipHKUST Undergraduate Admissions
- Based in
- Hong Kong SAR / Mainland ChinaEnglish & Mandarin
Education
-
—
Expected
Hong Kong SAR
The Hong Kong University of Science and Technology
BEng in Artificial Intelligence · Department of Computer Science and Engineering
AwardHKUST Full Scholarship, Undergraduate Admissions
Coursework: Data Structures · Design and Analysis of Algorithms
Experience
-
—
Internship
Beijing, China
Software Engineer Intern
Beijing Zhongkexunbo Communication Technology Co., Ltd.
Python · pandas · Dify · DeepSeek
- Cleaned and reshaped database CSV exports in Python (pandas), then audited the downstream SMS template library to find why near-duplicate templates kept accumulating.
- Built a Dify workflow that drives DeepSeek to extract template variables from message samples and return them in one fixed JSON schema.
- Fed that structured JSON back to the downstream service so templates with matching variables collapsed into one: a 90–95% reduction (10:1 to 20:1 merge), replacing a manual review pass per batch.
Projects
-
—
Award
One Thing — Belonging Memories
Third Prize, 2025 Mobile Application Innovation Competition (North China Region)
iOS and visionOS application · Team of three
Swift · SwiftUI · ARKit · RealityKit · Object Capture · OpenAI GPT-4o
- Owned the Apple-native 3D capture layer: wrapped ARKit and RealityKit Object Capture behind a narrow scan-to-model interface, so feature code never handled capture sessions or reconstruction state.
- Reused the stored models on Apple Vision Pro with a visionOS scene that presents the same captures in 3D.
- Integrated GPT-4o in a vision-plus-text step that turns a captured object and the user’s draft notes into a structured narrative, behind a fixed JSON contract.
- Built the SwiftUI interface and core application logic in Swift.
-
—
Coursework
Artificial Intelligence Algorithms Implementation
Introduction to Artificial Intelligence coursework
Python · NumPy · PyTorch
- Implemented Naive Bayes, a multilayer perceptron, and a convolutional neural network from scratch in NumPy, using matrix and vector operations only for training and prediction.
- Fine-tuned a pretrained LLM on a question-answer dataset as a binary True/False judge, then measured it on held-out pairs.
- Compared the from-scratch models with the fine-tuned LLM on identical splits to separate implementation behavior from pretrained priors.
Technical skills
- Languages
- Python, C++, Swift, Java
- AI & LLM
- PyTorch, scikit-learn, pandas, NumPy, Dify, DeepSeek / OpenAI API
- Apple platforms
- SwiftUI, ARKit, RealityKit, Object Capture, visionOS
- Tools
- Xcode, Git, VS Code, LaTeX