IPO audit foundation → TRON finance architecture → AI-native agent system design. Currently architecting personal and enterprise automation with Hermes Agent — a multi-model orchestration platform. Intelligent routing across Claude, ChatGPT, Qwen, Kimi, Grok, DeepSeek. Turning LLMs into production workflows: task routing, tool execution, human-in-the-loop approval, and autonomous operations at the application layer.
Capability Stack / Layer 01–04
IPO-grade audit discipline, custody architecture, and compliance controls. Every action traceable, every exception documented. The foundation everything else rests on.
Multi-chain operations across ETH, BTC, TRON, SOL. CEX wallet architecture (cold/warm/hot), on-chain forensics, DeFi settlement, custodian coordination.
LLM-assisted payment review pipelines, human-in-the-loop approval logic, risk trigger design, quantifiable interception rates. Automation with an audit trail.
Multi-agent orchestration, guardrail layer design, tool execution safety, structured human checkpoints. Systems designed for irreversible decisions.
System Literacy / Agent Design
Irreversibility thinking — every on-chain payment is permanent; agent tool calls must be designed with the same discipline
Audit trail culture — every action logged, every exception documented, every decision traceable to a human approver
Multi-party coordination — custodian, compliance, and legal handoffs map directly to multi-agent tool call boundaries
Multi-model intelligent routing — Claude, ChatGPT, Qwen, Kimi, Grok, DeepSeek — task-tiered dispatch with cost-performance optimization across reasoning, coding, and orchestration workloads
AI application layer architecture — turning LLM capabilities into production-grade workflows: task routing, tool execution, human-in-the-loop approval, autonomous operations
Enterprise automation — from personal AI assistant to company-wide automated pipelines spanning financial review, cross-department coordination, and knowledge management
Experience / 5 Chapters
Progression from financial execution to architecture ownership, with increasing responsibility at each layer. Each chapter compound-adds to the next.
Multi-agent orchestration with role-based dispatch -- separate agent types for planning, execution, and review, each with isolated context and bounded tool access.
Deterministic supervisor layer -- every agent action passes through pre-flight validation (budget check, scope gate, safety guardrail) before execution, with structured error handling and no silent fallbacks.
Three-tier human-in-the-loop approval system -- auto-pass for low-risk, review-required for medium-risk, hard-approval for irreversible actions, routed via Telegram inline UI.
Knowledge plane integration -- structured decision logging cross-referenced in a personal knowledge vault; every outcome traceable to a decision record.
Autonomous operation loops -- scheduled background research, token-cost monitoring, and system health checks running on cron with alerting.
Cost-aware model routing -- tasks dispatched to the optimal LLM provider and model based on complexity, budget, and capability.
Improved group-wide financial operations through automation, data restructuring, and process redesign.
Integrated AI Agent concepts into payment review workflows with human-in-the-loop approval logic.
Restructured financial data architecture for multi-chain ledger consistency.
Coordinated parallel cross-functional initiatives across Finance, Tech, Compliance, Legal.
Prepared financial architecture diagnostics and decision materials for senior management.
Supported throughput growth through workflow redesign and resource optimization.
Led 15-FTE multi-functional team (AP/AR, Operations, Treasury, Liquidity, Financial Reporting).
Owned end-to-end workflows for Blockchain, DeFi, and Custody verticals.
Managed CEX warm-wallet daily flow, multi-chain cash flow planning, and liquidity allocation.
Built Poloniex P&L reporting framework and settled DeFi investment yield to general ledger.
Designed market-making team master/sub-account architecture with API isolation and position boundaries.
Represented finance on exchange compliance queries; on-chain forensics to Legal & Compliance.
Executed group-wide multi-chain payments across ETH, BTC, TRON, SOL and cross-chain bridges.
Performed DEX token swap operations and cross-chain asset transfers for treasury needs.
4-dimension pre-flight validation: address, amount, network, memo.
Monitored on-chain broadcast and TxID confirmation; real-time anomaly escalation.
Led development management of enterprise ERM system from requirements gathering to delivery.
Managed VR/AR and web-based project development — virtual tours, mini-games, interactive experiences.
Liaison between production and clients — translated demand into delivery roadmaps.
End-to-end management of NFT and Metaverse projects: artwork, production, community.
Managed IPO projects end-to-end ensuring on-time filing — SEC Edgar & HKEX Form A1/5A.
Led cross-functional teams across professional parties; ensured effective stakeholder communication.
Monitored project progress, identified risks, and implemented mitigation strategies.
Planned and executed multi-industry audit and review engagements.
Drafted and reviewed prospectuses, annual reports, earnings announcements.
Executed special audits and annual audit due diligence across sectors.
Drafted and discussed regulatory responses to HKEX and SFC inquiries.
Selected Engagements
I don't just build agents. I build agents that can be trusted with irreversible decisions.
From a payment specialist who has executed hundreds of irreversible on-chain transactions
Capabilities / 7 Modules
Focused expertise at the intersection of financial control, Web3 execution, risk awareness, and practical automation.
Cold/warm/hot wallet design, wBTC Mint/Redeem protocol, custodian coordination, exception handling.
LLM Agent in payment approval pipelines, human-in-the-loop risk triggers, quantifiable interception rate.
Technical opinion documents for Legal & Compliance casework, exchange query resolution.
Group-level multi-department parallel projects, governance frameworks, auditable deliverables.
HKEX & SFC compliance, prospectus drafting, due diligence, response letters.
Direct delivery to C-suite, data-driven, outcome metrics and risk assessment only.
Role-based multi-agent orchestration with deterministic execution control, three-tier human-in-the-loop approval gates, pre-flight validation guardrails, structured decision logging, autonomous cron operations, and cost-aware LLM routing.
Philosophy / 5 Principles
01
Scale throughput without headcount. Skills reshuffling and role redesign unlock latent capacity.
"Zero headcount, doubled output."
02
Strip manually-stitched workflows. Scripts seal off human-error risk and deliver closed-loop systems.
"Humans tire. Code doesn't."
03
Delegate decision rights to team leads through standardized process and clear authorization.
"Decentralization is real resilience."
04
Define value by how you contribute, not what you're called. Builder constructs scalable infrastructure. Sweeper removes friction. Grower drives PMF. One person spans all three.
"A title describes your reporting line. Your archetype describes your impact."
05
Build systems that outlast individual tools. The agent framework persists beyond any single LLM provider or API. Invest in architecture, not point solutions.
"The framework outlasts the tool."
Education & Skills
University
Hong Kong Metropolitan University
Bachelor of Business Administration
Languages
Cantonese (native) · English (professional) · Mandarin (professional)