Halley

OpenSource

• MCPAdvisor 2025.02 Demo SourceCode

  MCP ecosystem expanding rapidly but with fragmented installation and high discovery costs — developed MCPAdvisor to manage MCPs via natural language: automatically retrieve, evaluate, and install the most suitable MCP Servers, solving AI Coding toolchain fragmentation pain points.

  Won 2nd Place at 2025 OceanBase AI Hackathon (Team Leader).

• WritingHelper 2021.02 Demo SourceCode

  I realized that there was a lack of tools for writing authentic IELTS phrases on the market, so I developed WritingHelper.
  Built on Node.js using a trie tree and responsibility chain pattern with LSP APIs for phrase autocompletion, Chinese translation, timing and counting.

  15,000+ downloads since launch.

Work Project

• Ant Group · Ant Insurance FinTech 2024.06 - present

  Responsible for the AI Agent application of "Insurance Chaxha" on Alipay — China's first insurance product with full-category Agent interpretation capability, serving **8 million MAU**. Also leading the general intelligent agent system and MCP-related work.

  • Insurance Chaxha · General Agent System 说明

    Addressing siloed Agent development and lack of reusable infrastructure — designed Agent Runtime-Executor-SessionPool three-layer general intelligent agent architecture, specifically implementing Claude Code-like general Agent Loop based on LangGraph, supporting Alipay Insurance Chaxha (8 million MAU) and innovative page AIGC real-time visual interaction.

    Implemented Harness Engineering (Prompt-Test-Verify loop) to achieve 100% AI-delivered requirements; drove full-process efficiency from design specs to frontend components via AI Coding workflow, establishing a replicable AI quality system. Project published as best practice in Alibaba Cloud Developer.

  • MCP Integration & Domain Agent Framework 说明

    Over 100 existing HTTP/HSF services could not be directly consumed by Agents — led MCPBridge construction to MCP-ize them, designed Ant Insurance domain MCP Server, built cross-business reusable toolchain based on OneAgent + MCPs paradigm, significantly reducing new business Agent onboarding cycles.

• Taotian Group · Tao Factory 2021.06 - 2024.06

  Participated in industry-finance integration and led "Money-Bill-Invoice Integration" engineering; later co-led the pre-sales AI module from scratch with mentor, covering RAG knowledge base, Q&A, recommendation and other intelligent chains.

  • Knowledge Base Pipeline 说明

    Low answer hit rate and fragmented knowledge sources — built multi-source fusion RAG knowledge base pipeline (multi-path retrieval + LLM reranking + generation), integrating merchant, operator, algorithm recommendation, and missed-hit knowledge sources, establishing scalable effective answer production loop, supporting Tao Factory full-category pre-sales and after-sales intelligent Q&A.

  • Money-Bill-Invoice Integration 说明

    Led the automation and integration of “settlement”, “billing” and “supplier invoicing” across 10+ billing services with data volume reaching tens of billions. Built a DSL for faster onboarding, applied DDD, SPI and responsibility-tree patterns. Reduced chimney-style development cycles to one week. Featured in “Alibaba Technology” publication.

• Alibaba · Taote Business Department 2021.06 - 2022.10

  Responsible for replenishment & inventory planning in supply chain, ensuring goods reach shelves at the right quantity, time and place to meet order demand.

  • Replenishment Pipeline Reconstruction

    As the 1.5M-item product pool grew, the entire pipeline from data integration failed nearly every week. Rebuilt from scratch: pool reduction, pipeline optimization, and proactive failure monitoring.

    Selected search engine HA3 with SARO, Swift, and Flink. Implemented HTTP client and ORM from scratch; applied async, concurrency, and slicing optimizations along with strategy/redundancy business-level tuning.

    Reduced work orders and failures immediately; shortened recalculation time by 93%. Awarded “Making Users Happy” commendation.