Profile
Focus: AI application engineering and production delivery, backed by 7 years of
full-stack data and hands-on LLMOps experience.
Built deep expertise in enterprise data infrastructure, leading Prefect-based
secure orchestration and full-scope operations for million-scale offline workloads.
On top of that foundation, I systematically integrated LLM and agent systems
into traditional database querying, metric monitoring, and self-service reporting workflows.
My approach is simple: build on a solid data foundation, then use modern AI tooling to
multiply delivery efficiency.
Key Highlights
80+
🌐 Integrated 80+ domestic and global models, reducing the team's
average monthly cost to $20
70%+
🤖 Built an AI self-service data assistant that automated over 70% of
recurring data requests
1.3M+
⚙️ Led orchestration for 3 years with 1.3M+ monthly task runs,
automating SQL execution, access control, and log cleanup
Top 5%
🏆 Featured in official newsletters
twice, with public recognition from the original author tjbck and core expert Classic298
Data Team Productivity Systems
🤖 AI Data Request & Operations Assistant
Deep AI + DB Integration
70%+ request automation
- SQL automation assistance: LLMs interpret requests, generate SQL, and run built-in safety checks across 70% of common query scenarios.
- Semantic schema discovery: AI ranks candidate tables using metadata and historical usage, reducing time spent locating data sources.
- Performance risk prediction: SQL risk analysis and optimization guidance reduce the impact of heavy queries on production databases.
⚙️ Prefect Automation Orchestration Platform
- Automation architecture: Orchestrated 70+ core data flows with Prefect, supporting 1.3M+ monthly runs in a 24x7 unattended environment.
- Operations automation: Connected SQL auditing, access provisioning, log cleanup, and 20+ operational actions into approval-to-execution workflows.
- AI-assisted data delivery: Built an end-to-end pipeline from request classification to routing, generation, and distribution, covering roughly 70% of task interpretation work.
- High-reliability scheduling core: A single Python runtime coordinates complex multi-source heterogeneous database workflows with failover support, keeping success rates above 97%.
- Full-path monitoring: Integrated enterprise WeChat APIs for second-level incident alerts and automated log diagnostics.
🛡️ Enterprise Unified AI Gateway
- Extreme cost efficiency: Aggregated Copilot, Gemini, CLI Proxy, ModelScope, OpenRouter, and DeepSeek reverse-engineered access to cover 80+ mainstream models, while keeping heavy daily usage near $20/month through routing and quota control.
- Infrastructure resilience: Mixed deployment architecture with automated IP rotation to keep the team's AI tooling stable under provider-side controls.
Open-Source Ecosystem & AI Tooling 2024.12 - Present |
Core Ecosystem Developer
⭐ 191 Stars
(recognized by original author tjbck and core expert Classic298)
🔥
Official newsletter featured twice
in community
highlights
📂 Actively maintained
OpenWebUI community: 👁 21.8W+ views | ⬇ 20,836 downloads | 👥 794+ followers
- Deep GitHub Copilot integration: Exposes native Copilot capabilities and uses Actions for instant Excel and HTML rendering.
- Rich media rendering: Transforms content into mind maps and live charts for better reasoning visibility.
- Asynchronous context compression: Compresses conversations by semantic importance to cut token cost and improve long-context stability.
- Response normalization: Cleans noisy or malformed model output in real time for higher readability.
⭐ 38 Stars
(starred by core expert Classic298)
Released in 2026.01
- Built in pure JavaScript, with Spotlight-style global shortcuts and millisecond fuzzy search.
- An embedded AI agent turns natural language into structured prompts and
automatically renders
into user-friendly UI controls such as dropdowns and date pickers.
⭐ 75 Stars · Fork
26
Fixed status leakage and parsing failures caused by DeepSeek protocol changes, added
citation parsing for R1 web search, and shipped multi-arch Docker
images. It became a preferred replacement in the community, demonstrating strong
protocol reverse-engineering and engineering adaptation capability.
PyPI 16.1K+ Downloads
⭐ 31 Stars
(starred by original author tjbck)
A stateful Python client SDK for LLM orchestration workflows, supporting async/sync modes,
multimodal inputs, and tool calling. It is already used as a core building block in internal
automated reasoning pipelines.
Experience
Shanghai M&G Colipu Office Supplies Co., Ltd.
Senior Data Engineer / AI Productivity Lead
2021.11 - Present
- Scheduling foundation: Built the enterprise data orchestration stack from scratch, supporting company-wide data flows with a 97%+ success rate.
- AI-powered operations: Integrated LLM capabilities into daily database operations and approval flows, automating about 70% of repetitive requests.
- Developer productivity: Helped the team move from manual ETL toward AI-assisted development, greatly improving individual analyst throughput.
- Model ecosystem enablement: Deployed an enterprise LLM gateway, retrieval-enhanced pipelines, and an evolving prompt template library.
Nanji E-Commerce (Shanghai) Co., Ltd.
Data Analyst
2021.06 - 2021.11
- Built an automated e-commerce data monitoring system and delivered performance dashboards in Power BI.
Shanghai Tongjie Data Technology Co., Ltd.
Data Engineer
2019.03 - 2021.06
- Developed an automated market research report generation system that significantly improved report delivery efficiency.