Applied AI systems — from research to reliable agents.

I build systems that retrieve context, orchestrate tools, and turn engineering claims into verifiable outcomes.

AI Agents · Developer Tools · Trustworthy Automation · Applied Research

从应用研究到可靠 Agent 系统，关注上下文、受控执行与可验证结果。

Email · Selected systems · Research

This is a map of shared capabilities and engineering principles, not a claim that every project is part of one runtime system.

• AskAny — A Chinese-optimized RAG assistant with hybrid retrieval, reranking, tool-call traces, and source provenance.

• Skills — A catalog of small, composable skills for decision clarification, read-only review, and review-gated delivery.
• AgentRepoRouter — Repo-aware routing across repositories, project guidance, and native coding CLI backends.
• DocMate — Documentation QA that cross-checks code, reports gaps, and can prepare isolated documentation repairs.

• AI-Codereview-Gitlab-Opencode — Multi-provider code review with a durable queue, disposable worktrees, multiple agent backends, and provider-native delivery receipts.

• Reviewworthy — Maintainer-first workflows for human-owned AI-assisted contributions, with policy gates and verification evidence.
• Git Evidence — Provider-neutral engineering activity reports with source-linked claims and explicit coverage boundaries.

• Link every engineering claim to verifiable evidence.
• Give automation explicit authority boundaries.
• Make automated actions visible and reviewable.
• Use platform-native capabilities before building extra layers.
• Keep humans responsible for final decisions and outcomes.

• YOLO-World — Deployment and inference fixes covering operator export, optional text input, batched NMS, and detection counts (#174, #211, #216, #297).
• TorchEEG — Tensor-safety and visualization fixes plus FACED dataset support (#106, #107, #108).
• sk2torch — Dtype compatibility for support-vector inference (#2).
• gitlab-mcp — GitHub-style toolsets and configurable tool filtering (#345).

• CVPR 2022 · OakInk: A Large-scale Knowledge Repository for Understanding Hand-Object Interaction
• ICASSP 2024 · An Attention-Enhanced Retentive Broad Learning System for Subject-Generic Emotion Recognition from EEG Signals

My research background spans computer vision and neural signal modeling; my current work explores how those foundations can support reliable agent systems.

Generated weekly from public GitHub data with GitHub Profile Summary Cards.

For research, open-source collaboration, or developer-tooling conversations: wufeii.sjtu@gmail.com.