Ray Tse谢梓睿 Ray Tse

Java 17 · Spring Boot 3 · Spring Cloud Alibaba · Nacos · OpenFeign · MyBatis-Plus · MySQL · Redis — cross-border payments microservice platform; owner of the operations console service.Java 17 · Spring Boot 3 · Spring Cloud Alibaba · Nacos · OpenFeign · MyBatis-Plus · MySQL · Redis —— 跨境支付微服务平台，负责运营平台（console）服务。

• Delivered the operations console microservice from zero (Spring Boot 3, Spring Cloud Alibaba, Nacos): 80+ REST endpoints across 10+ business modules — merchants, KYB, risk, order review, reconciliation.从 0 到 1 交付运营平台微服务（Spring Boot 3 + Spring Cloud Alibaba / Nacos）：80+ 个 REST 接口，覆盖商户、KYB、风控、订单审核、对账等 10+ 个业务模块。
• Connected 5 downstream services (user, account, order, withdrawal, config) through OpenFeign, relaying login state with a RequestInterceptor for unified cross-service auth.基于 OpenFeign + RequestInterceptor 打通用户、账户、订单、提现、配置 5 个下游服务，登录态跨服务透传，统一鉴权。
• Shipped Redis-backed distributed sessions and Spring AOP data scoping with button-level RBAC, so sales staff see only their own merchants.落地 Redis 分布式会话，以及基于 Spring AOP 的数据权限 + 按钮级 RBAC，销售人员只能访问名下商户。
• Built the MyBatis-Plus data layer (MySQL schema, shared pagination and responses, audit-field auto-fill) and the EasyExcel reconciliation export; fixed boot failures from Spring Boot 3 auto-configuration changes and a bean circular dependency.搭建 MyBatis-Plus 数据层（MySQL 表结构、统一分页与响应、审计字段自动填充）与 EasyExcel 对账导出；解决 Spring Boot 3 自动配置失效与 Bean 循环依赖导致的启动失败。

Open source · macOS desktop agent · TypeScript, Node.js, Electron, React, Swift开源 · macOS 桌面 Agent · TypeScript、Node.js、Electron、React、Swift

• Built solo a macOS Computer Use agent: an agent loop with multi-step Function Calling that reads the UI from multimodal screenshots and the Accessibility tree and acts via file, shell and desktop tools.独立设计并实现面向 macOS 的 Computer Use Agent：基于 Agent Loop + Function Calling 的多步工具调用，结合多模态截图与 Accessibility 树理解界面，执行文件、Shell 与桌面点击 / 滚动 / 输入；支持流式输出与 JSONL 会话持久化，Agent runtime 与 Electron 前端通过 JSONL 协议解耦。
• Context engineering: automatic and manual Context Compaction against the model's context window, with a provenance rule that keeps summaries from turning tool output into user goals; token-budgeted long-term memory and screen-activity memory injected each turn and traceable to source frames.上下文工程：按模型 context window 自动 / 手动 Context Compaction，以 provenance 规则防止摘要把工具输出误当成用户目标；基于 token budget 的长期记忆与屏幕活动记忆每轮注入，并可溯源到源帧。
• Safety and human-in-the-loop: Seatbelt-sandboxed Bash; least-privilege approval for each file change, host command and desktop action; memory and tool output injected as untrusted data against Prompt Injection; cancellation voids pending approvals and flags uncertain side effects.安全与 Human-in-the-loop：Seatbelt 沙箱执行 Bash，文件修改、宿主命令与桌面操作按最小授权逐次审批；记忆与工具输出作为不可信数据注入以防御 Prompt Injection；取消时作废待审批项并标记不确定副作用，保障中断恢复。
• Evals and observability: five agent eval suites (tasks, file / Bash security, desktop security, memory, compaction) with deterministic checks plus LLM-as-Judge grading (latest 32/34, 5/5, 8/8, 9/9, 12/12); turn-level tracing of inference, tokens, tool calls and approvals separates model, tool and runtime failures.评测与可观测性：搭建 5 套 Agent 评测（通用任务、文件 / Bash 安全、桌面安全、记忆、Compaction），确定性检查 + LLM-as-Judge 语义评分（最近一次 32/34、5/5、8/8、9/9、12/12）；Turn 级 Tracing 记录推理、Token 用量、工具调用与审批，用于区分模型、工具与 runtime 故障。