• Game development: Experienced with performance testing and analysis workflows used in game projects, from data collection and metrics to diagnostics and tooling.
• Performance analysis: Work with frame timings, CPU/GPU/FPS metrics, memory snapshots, and ADB data to compare trends, study correlations, and produce visual reports.
• Agent development: Explore agent patterns and modular combinations of tools, memory, RAG, MCP, and context engineering.
• Automation: Turn repetitive workflows into scripts, data pipelines, CLIs, desktop applications, and bots, with an emphasis on diagnostics and recoverability.

A local-first desktop resource manager for Codex, Claude Code, Cursor, Windsurf, and OpenCode. It safely synchronizes Skills, MCP servers, Rules, Prompts, and Plugins through a private Git repository controlled by the user.

• A shared Python core powers the desktop UI, CLI, and MCP server.
• Every write is planned before execution, with ownership tracking, target locks, backups, validation, and rollback.
• MCP credentials are replaced with placeholders before resources enter version control.
• The project ships as a Windows installer with environment diagnostics, CI, release notes, and user documentation.

A modular Python agent framework for research and experimentation. Reasoning, tools, memory, retrieval, and context management are designed as independently reusable components.

• Six agent patterns: Simple, Function Calling, ReAct, Plan-and-Solve, Reflection, and Context-Aware.
• Working, episodic, semantic, and perceptual memory, plus a RAG pipeline with query expansion, HyDE, and reranking.
• MCP tool integration, GSSC context engineering, and a unified interface for cloud and local OpenAI-compatible models.
• End-to-end examples for PDF learning, codebase maintenance, document generation, and search.

A self-hosted, declarative platform for deploying, operating, and evaluating multi-channel AI agents. It uses versioned BotSpec configurations to compose each agent while sharing contracts, adapters, operations, and evaluation infrastructure.

• Per-instance selection of a native runtime, LangGraph, Codex, and other backends.
• Feishu and QQ / OneBot integrations with composable tools, MCP services, RAG, memory, and private wikis.
• Identity, role-based access, task isolation, credential boundaries, diagnostics, and logs.
• A React / FastAPI operations console, Linux / WSL deployment, and BFCL, GAIA, and IFEval evaluation support.

• Agents: LLM Agents · ReAct · Plan-and-Solve · Reflection · Function Calling · LangGraph · RAG · MCP · ACP · Context Engineering
• Applications: Python 3.10+ · FastAPI · Pydantic · React · SSE · WebSocket · Desktop GUI · CLI
• Data & performance: Qdrant · Neo4j · SQLite · pandas · NumPy · Matplotlib · OpenPyXL · Unreal Insights · ADB
• Engineering: pytest · Ruff · mypy · Git · CI · Docker · systemd · Windows · Linux / WSL