Hello! I'm a third-year Software Engineering student @ North University of China. 😘 Focused on Java microservices & middleware, LLM runtime internals, and Spec-driven Coding under engineering harnesses, prioritizing maintainability and system boundaries. 💫

• Self-developed highlights:
    🔹 Maintaining wolfcha (AI Werewolf social deduction with dynamic multi-agent persona scheduling & modern UI)
    🔹 Porting & maintaining pane (OpenUsage Windows tray port, real-time quota dashboard for 15+ AI providers)
    🔹 Co-developing AgentCFO (DAO treasury policy compliance & on-chain approval automation, hackathon finalist)
    🔹 Building simple-ai-code-helper (Spring Boot + LangChain4j + MCP AI coding assistant)
    🔹 Customizing fork-Firefly (Astro-based blog & digital garden fork with automated agent publishing pipelines)
• Harness Engineering & TTA Toolchain: Turning spec-driven experience into reusable engineering assets; open-sourcing the TTA skill fleet — tta-init (governance baseline for agent maintenance) · tta-html (interactive HTML showcase & slides) · tta-visual (high-fidelity generative visual assets) · tta-tone (natural humanized copywriting & review) · tta-ppt (presentation slide generators) · tta-frontend (production-ready web templates); plus cross-harness sync via harness-sync.
• Open Source & Fullstack Engineering: Re-architected & shipped sky-out-ai (AI-enhanced takeout platform, 3-tier runnable, customer service RAG); deep-dived fork-yu-rpc (Java RPC framework) & fork-yu-picture (collaborative cloud gallery); actively maintaining fullstack and community repos: ResumeWise · fork-Firefly · fork-daybook · fork-mianshiya · fork-sql-mother etc.

Outside tech: Road cycling enthusiast 🚲, following Grand Tours & Monuments; writing blogs, cultivating digital gardens, and staying curious beyond the screen.

Frontend & Fullstack

Java / Spring

Middleware & microservices

AI / agents

Systems / inference — competition proof

Python

DevOps

2026 全国大学生计算机系统能力大赛 · 智能计算创新设计赛（先导杯）赛题 1 · Team 翻斗花园

Proof split: official best anchor is 87.7839 / 100 · #26 / 132 · SLA 0 · precision 0; our own best anchor is 87.6933 / 100, with 4k–8k / 8k–16k / 16k–32k throughput 20.39 / 18.29 / 14.61 tok/s and precision 0.

Workload: raise long-context Qwen throughput on a fixed domestic DCU at concurrency=1, under TTFT/TPOT P99 SLA (input mix 4k–8k / 8k–16k / 16k–32k).

Validated recipe: fused shared-gate ON · LPK stages=1 · Gather-FA ≤16K · rocBLAS + TunableOp · LPK prefetch. vs official baseline smoke: TTFT P99 −61%–87% · TPOT P99 ≈ −35% · throughput +7%–24%.

2026 书生国智科探挑战赛暨飞翔杯 AI Agent/Skills 开发大赛 · Shanghai AI Lab × Biren · Track 5 模型与算子 · Team 翻斗花园

Live board: public NS64 rel-L2 0.035115 · tag dualview_r2 · report v9 · Spectral idle 3.811 / 8.054 / 29.560 ms @64/128/256 · worst rel ≈ 2.17e-7 (≤1e-4) · ranking pending

Problem: ship a Biren-native Spectral Convolution (SUPA / Extension) and reuse it inside ≥4-layer FNO-NS on the public 64×64 Navier–Stokes set (1000/128) — Agent/Skills log required (~15%).

Evidence: formal Spectral idle frozen; chain CPU↔SUPA <1e-4; Pred/GT viz on public NS64; Agent log 35+ reviewed segments with abort/NO_SIGNAL discipline (no silent promote).

A treasury assistant for DAO operators 🤠 powered by Cobo Agentic Wallet. Automates payout proposals, policy compliance checks, and multi-party on-chain approvals, replacing fragmented spreadsheets and manual transfers. Hackathon finalist prototype.

• Track: Cobo · Agentic Economy × CAW
• Role: Frontend lead; designed and built the web landing page and operator console
• Flow: Contribution records & budget rules → automated payout plan → deterministic policy checks → multi-party human approval → on-chain execution & audit log
• Proof: Two end-to-end verified settlements on Sepolia testnet (external payment + internal transfer)
• Stack: Next.js · TypeScript · FastAPI · Cobo CAW

Systematically organizing microservice governance (Spring Cloud Alibaba, Nacos, Sentinel, Gateway) and middleware reliability. Building a distributed delay-delivery system focused on high-concurrency peak shaving, zero message loss, and cluster failover.

Spring Cloud RocketMQ Nacos Sentinel

Exploring LLM gateway routing, stateful Agent Loops with checkpoints, MCP tool runtimes, and hybrid-retrieval RAG. Customizing enterprise Dify knowledge bases and workflows, alongside Subagent orchestration, Golden Datasets, and LLM-as-Judge automated evals.

Agent Loop Checkpoint Dify Plugins Evals

Evolving from Prompt → Context → Harness → Loop Engineering, using discipline to shrink human-agent gaps and force clearer architecture thinking. Practicing Spec-Kit and OpenSpec upfront contracts to align cognition, connecting dual-track iteration with MCP-assisted E2E testing to catch broken business paths early.

Spec-Kit OpenSpec Dual-Track E2E MCP

Diving into DSH plugin internals—contributing an enhancement plugin to the official community while developing another workflow plugin from scratch. Syncing multi-agent configs with custom harness-sync, and distilling practical workflows into the TTA skill fleet toward spec-driven development.

DSH Plugins harness-sync TTA Skills

Tinkering with Grokbot interactions, proactive listening, and automated monitoring workflows. Moving from idea to prototype (PM → Builder), weaving persistent memory, proactive triggers, and Generative UI into real application scenarios.

Grokbot PM➔Builder Active Loop Generative UI

Connecting mobile apps and mini-programs with international payments (Stripe). Letting information drive development while filling gaps with AI, validating real demand and indie shipping paths with lean MVPs.

Mobile App Mini Program Global Stripe MVP