SYSTEM ARCHITECTURE · AGENT ENGINEERING · PRODUCT
Unbound by tech stacks, building ideas into usable products.
I’m Fish. Trained in physics and scientific computing at the master’s level, I bring over a decade of software development and systems architecture experience. Today I focus on AI agents, data applications, and product engineering across enterprise systems, quantitative research, cross-border commerce, and Web3D—clarifying domain boundaries and turning complex needs into dependable, evolving systems.
Journal
Latest Writing
Node Monitoring and Timeline Replay for the Smart Cockpit Agent
Addressing the practical debugging question of whether an agent actually called downstream APIs, using an ordered event stream for live node monitoring and frontend prefix replay while handling loop span pairing, SSE subscription races, and thread limits.
Orchestrating Agent Workflows with State Graphs in the Smart Cockpit
Starting from why a single prompt struggles with multi-step business logic, this article walks through Graph-based state models, tool loops, multi-specialist collaboration, and how customer-service workflows connect to Coding Agents and cybernetics.
Business Ledger and Agent Interaction Design for 3D Smart Parking
Reusing the Enterprise AI Cockpit backend, a reproducible year of synthetic parking records, server-side role checks, a versioned POI road graph, and AG-UI events connect a fullscreen 3D campus to parking recommendations, work orders, analytics, and voice/vision controls.
Labs
AI and 3D Engineering Projects
Interactive online deployments spanning a 3D campus twin, quantitative research, cross-border product analytics, and an enterprise AI workbench, built to validate system architecture in concrete workflows.
Running3D Smart Parking
A fullscreen Babylon.js 3D campus with on-demand overlays for bay occupancy, route guidance, alert work orders, and operations analytics across desktop and mobile.
RunningAI Quant Research Cockpit
Combines market data, point-in-time financials, news sentiment, factor computation, and walk-forward backtests with timeline-level signal and cost checks.
RunningCross-border Trend Report
Aggregates multi-source product listings, search interest, and exchange rates to normalize categories, estimate fulfillment margins, and produce source-linked daily reports.
RunningEnterprise AI Cockpit
Integrates hybrid knowledge retrieval, MCP report queries, state-graph workflows, and node execution replay for source-cited Q&A and trace inspection.