English | 简体中文

AI Agent & Quant Systems Builder · Big Data Engineer

I build governed AI Agent runtimes, quantitative research systems, and end-to-end AI workflows. My approach connects models to real data and tools, then adds durable state, evaluation, permissions, human approval, and runtime verification so the result is useful beyond a prompt demo.

This work is backed by 9 years of big-data development experience across PB-scale offline data warehouses, large-scale streaming, workflow governance, data quality, and performance optimization.

中文主页 · Email

A governed quantitative-research Agent runtime that turns open-ended research goals into durable, reviewable missions.

• Persists goals, plans, steps, evidence, budgets, and completion conditions
• Combines MCTS and MAP-Elites for diverse strategy-code exploration
• Uses red-team stress tests, regime attribution, and structured negative constraints
• Keeps research, backtesting, Paper trading, and real effects behind explicit control boundaries

An ARC-AGI research system spanning ARC-AGI-1/2 program synthesis and ARC-AGI-3 interactive-Agent experiments.

• Grid-transformation DSLs, candidate generation, exact-match validation, and restricted code execution
• Visual-state abstraction, action history, skill routing, and trajectory-based evaluation
• Separates local diagnostics from official benchmark results and retains reproducible evidence

Planning · Tool Calling · MCP · JSON Schema · State Machines · Memory · Evaluation · Human Approval · Recovery · Audit

• Alpha — Open-source A-share research system with trusted market-data ingestion, reproducible workflows, CI, and a public release
• StockPro — Real-time A-share research and monitoring platform with data quality, strategy lifecycle, backtesting, Paper trading, and operational checks
• QuantBase — Research workbench for real market data, Backtrader validation, Paper trading, signal audit, and risk-first development
• BitPro · Private Product — Digital-asset research platform covering exchange data, strategy versions, asynchronous backtests, simulation, controlled execution, and monitoring

I treat market-data quality, reproducible computation, costs, fills, and audit trails as prerequisites. Strategy research is presented as research evidence—not as unverified return claims.

• Zora · Private Product — AI animation workspace connecting story, characters, storyboards, video generation, voice, composition, quality review, and publishing
• FrameLab · Private Product — AI video platform integrating model APIs, ComfyUI/GPU workers, asynchronous jobs, storage, credits, moderation, testing, and deployment

I use Codex, Cursor, GLM, Grok, and other models according to their capabilities and limits. I remain responsible for business judgment, requirement decomposition, constraints, acceptance criteria, and verification through real data, automated tests, runtime logs, and user-visible outcomes.

A food-ingredient analysis and health-literacy mini program that I continuously operate and improve, covering data organization, product iteration, and promotion through WeChat Search.

A fishing-focused tide and weather mini program integrating time-series tide and weather data, backend services, and a mobile-facing product experience.

Both products are continuously operated and iterated—not one-off demos.

My big-data strengths cover dimensional and layered warehouse modeling, Flink/Kafka streaming, large-state checkpoint and memory diagnosis, scheduler migration, metric governance, cross-region synchronization, historical backfills, observability, and failure recovery.

Core stack: Flink · Kafka · Spark · Hive · Hadoop · HBase · Airflow · MySQL · PostgreSQL · ClickHouse · Python · Java · Scala

Employer source code, business data, and internal implementation details remain confidential. Private projects are described by capability and verified outcomes without exposing their repositories.

• Governed Agent runtimes with evidence, evaluation, memory, and safe tool use
• Real-data quantitative research with reproducible computation and Paper-first validation
• AI video and content-production systems with human review and observable delivery
• Reliable data foundations for batch, streaming, model, and Agent workloads