English · 简体中文

Electronic Information Engineering undergraduate at Shenzhen University · Class of 2027

I build reproducible machine-learning systems and the engineering tooling around them. My current focus is image and video quality assessment (IQA/VQA) end to end: data validation, model training, evaluation, interpretability, and a FastAPI + Nginx deployment.

• ML engineering — configuration-driven training, evaluation, checkpoints, and experiment artifacts
• Computer vision — image and video quality assessment with spatial and temporal modeling
• Serving & deployment — FastAPI services, batch inference, Docker/Podman, Nginx, and CI validation

A PyTorch platform for image and video quality assessment, covering data checks, model training, evaluation, and inference deployment.

• A unified IQAVQANet for image IQA and video VQA, using configurable ImageNet backbones and Transformer-based temporal fusion
• Media integrity checks that decode inputs before training and isolate unreadable samples
• Group-aware train/validation/test and K-fold splitting to reduce leakage from repeated reference content
• A hybrid MOS regression and pairwise-ranking objective for quality prediction
• Evaluation and interpretability workflows covering PLCC, SROCC, KROCC, RMSE, residual diagnostics, MOS-bin analysis, feature distributions, and Grad-CAM
• A browser workbench for uploading media, running selected checkpoints, viewing scores and history, and inspecting model interpretations
• SQLite-backed evaluation history for internal single-instance deployments, with a stateless storage mode for protected public deployments
• Docker/Podman deployment with Nginx, health checks, deployment profiles, GitHub Actions validation, and uv-managed environments

Codex skills for repository maintenance and skill authoring, with reusable workflows for documentation, GitHub Actions, repository hygiene, project layout, frontend/backend quality, PowerShell, and Docker/Podman deployment. The repository also includes skill validation and routing evaluations.

Reusable Codex skills for Python engineering and code review in medium and large repositories. They turn recurring decisions into practical review criteria for agent-generated changes.

The skills cover domain modeling, callable and caching decisions, scikit-learn workflows, file-processing CLIs, plotting, logging, media processing, and binary buffers. The repository also includes cross-platform project installation, package validation, routing datasets, and Codex skill discovery and activation evaluations.

Python · PyTorch · OpenCV · FastAPI · SQLite · Nginx · Docker/Podman · Makefile · GitHub Actions · Codex · Bash

Entry-level roles or internships in ML engineering, computer vision, or backend engineering.

I care about reproducibility, clear interfaces, and making ML workflows easier to run and maintain.

Reach me at icey08852@gmail.com.