I'm a Master of Data Science graduate from UBC Okanagan, focused on applied data science, AI-assisted analytics, computer vision workflows, and stakeholder-facing data products.
I like building systems that turn messy real-world information into structured evidence, visual analytics, and decision-ready tools — from zoning-bylaw rule extraction and GitHub AI-agent behavior analysis to interactive dashboards, Python packages, and vision-based workflow monitoring prototypes.
- Data analytics & visualization — dashboards, reporting workflows, exploratory analysis, and stakeholder-facing insights
- AI-assisted data systems — LLM/RAG-style extraction, verification workflows, document processing, and evidence review
- Computer vision & workflow monitoring — segmentation, object tracking, action recognition, transfer learning, and GUI-based model experimentation
- Python/R software projects — reusable packages, modular pipelines, testing, documentation, and reproducible analysis
| Project | What it shows |
|---|---|
| GitHub AI Agent Behavior Research | Empirical analysis of AI coding-agent pull requests, including data cleaning, feature engineering, quality modeling, failure-pattern analysis, repository clustering, community detection, and report-ready visualization. |
| Rule Extraction Demo Pages | Static demo site for a Green Metrics Technology capstone pipeline that converts zoning bylaw PDFs into visual blocks, evidence packs, Graph/RAG-style retrieval outputs, and LLM-ready structured rules. |
| Spotify Track Insights Explorer | Interactive dashboard for exploring genre, mood, popularity, track profiles, and similar-track discovery through a user-facing analytics interface. |
| Project | What it shows |
|---|---|
| Prompt Segmenter Model | GUI-based first-person video object segmentation and tracking pipeline, combining YOLO11-seg, SAM2-style segmentation routes, ByteTrack tracking, auto-labeling, clustering, and pseudo-label export. |
| ModelCraft Studio | GUI-first transfer learning studio for image classification experiments, with training, prediction, robustness evaluation, Grad-CAM inspection, custom model generation, and log-driven comparison views. |
| Egtea Gaze Training | Early workflow-monitoring experiments around egocentric video, action recognition, and temporal modeling. |
| Project | What it shows |
|---|---|
| Budget System Python Package | Published-style Python package for household finance management, with modular OOP design, subpackages for members, budget funds, and assets, CLI workflows, pandas summaries, and visualization-ready outputs. |
| bocvaletR | R package / R workflow development experience. |
| DATA551 Render Test | Dashboard deployment and reproducibility testing. |
Languages: Python, R, SQL, JavaScript
Data & analytics: pandas, NumPy, scikit-learn, statistical analysis, feature engineering, reporting, dashboarding
Visualization: Plotly/Dash, Matplotlib, NetworkX, dashboard design, graph/network visualization
AI / ML: PyTorch concepts, transfer learning, computer vision, segmentation, tracking, model evaluation
Document AI: PDF processing, rule extraction, LLM-assisted verification, Graph/RAG-style evidence organization
Software workflow: Git/GitHub, modular project structure, package development, testing, reproducible notebooks, CLI/GUI tooling
I'm especially interested in:
- applied analytics roles where data products support real decisions;
- AI-assisted document intelligence and verification workflows;
- computer vision systems for segmentation, tracking, and workflow monitoring;
- open-vocabulary / RGB-D-style visual understanding pipelines;
- building clean, reusable tools that make analysis easier for non-technical users.
- GitHub: github.com/Zihao-Sheng
- LinkedIn: linkedin.com/in/zihao-sheng-320b1632a
- Email: [email protected]