Delon XIE

Web3 & Financial Risk Expert: Hands-on experience managing 150K ETH Staking and $500M USD digital asset custody. Deep understanding of digital asset trading, risk control, and compliance (KYT/KYC) systems. Previous experience in P2P finance and mortgage loan risk control forms a complete fintech knowledge framework.

AI-Driven Full-Stack Architect: Profound insight into AI — AI is not a silver bullet but a collaborative partner requiring demand-driven, problem-oriented, architecture-adapted, and test-covered approaches to maximize value. Skilled at precisely identifying genuine pain points where AI can provide solutions within complex systems.

Full-Stack Architecture & Scale Experience: End-to-end architectural vision from exchange matching engines (Rust/Raft) to AI-powered quantitative stock selection (Qlib/LightGBM + LLM MCP), from low-level hardware (RFID encryption cards, PLC/RS232) to upper-layer applications (Vue/React), from real-time stream computing (Flink) to offline scheduling (Airflow/Crontab). Practical experience managing 60+ person teams and building financial-grade systems from scratch.

Deep Decomposition & Abstraction: Capable of systematically decomposing high-complexity technical challenges such as bytecode decompilers, MQTT protocols, and multi-market quantitative trading systems using first-principles thinking, fundamentally solving long-standing industry robustness issues through innovative architectures.

While leading AI empowerment projects at Avenir, my understanding of AI can be summarized in five keywords: Demand-Driven, Problem-Oriented, Architecture-Adapted, AI Collaboration, Test Coverage.

• Demand-Driven: Any AI capability introduction must originate from genuine business needs. In QuantByQlib, the dual LLM fallback architecture (Claude primary + DeepSeek fallback) was adopted because users genuinely needed high-quality stock analysis reports with guaranteed service stability and cost efficiency.
• Problem-Oriented: Define the problem clearly first, then choose the right technical approach. In PyRebuilderSharp, the core problem was traditional decompilers’ poor robustness — not “how to write more code with AI.”
• Architecture-Adapted: AI capabilities must integrate into the overall architecture design, not as isolated “black box” plugins. In QuantByQlib, RD-Agent factor discovery results are automatically injected into LightGBM strategies through standardized IC validation pipelines.
• AI Collaboration: AI is a powerful collaborator but can never replace human architectural judgment and domain insight. AI handles rapid prototyping, boundary condition completion, and repetitive code generation, while architectural decisions, problem decomposition, and quality control remain my responsibility.
• Test Coverage: Quality verification of AI-generated code is indispensable. I established a baseline coverage testing mechanism — in PyRebuilderSharp, 182 real .pyc files form a test suite ensuring no regression with every change.

My management philosophy can be summarized as a three-layer funnel model: Strategic Alignment → Engineering Efficiency → Talent Pipeline.

Strategic Alignment: Technology team goals must be highly aligned with business objectives. At Avenir Tech, I aligned AI empowerment engineering with the group’s cost reduction goals, directly driving server costs from $150K/month to $45K/month.

Engineering Efficiency: Through standardized development processes, automated toolchains, and AI-assisted development, continuously improving team delivery speed and code quality. Implemented Code Review systems, CI/CD pipelines, and unit test coverage thresholds, reducing production incidents by over 80%.

Talent Pipeline: Focus on cultivating team members’ technical depth and business understanding. Maintain weekly tech sharing sessions and one-on-one meetings, helping members develop personal growth paths. Among managed teams, 3 members grew into tech managers and 2 transitioned to architects.

Management Style: Results-oriented — no reports without quantitative metrics; Empowerment with trust — giving team members full autonomy and decision-making space while establishing clear accountability; Technical instinct — maintaining sensitivity to cutting-edge technology, spending 70% of energy on the R&D front line.