Sun Xinshan
+86 135 8591 7707|sunxinshan@163.com|github.com/sunxinshan
Summary
Thirteen years in software development, including six as a technical team lead and architecture owner, concentrated in financial risk control, real-time marketing and cross-border payments. I turn business requirements into architectures that can keep evolving, and have led LBS, the risk-control middle platform, the external vendor gateway and the marketing platform from proposal to production — while holding stability under load: LBS peaks at 4,000+ QPS and 200 million records a day, and the vendor gateway handles 1.5 million calls a day. Recipient of Ping An Bank's 2019 Best Architect (department) and 2020 Best Project Quality awards.
Core skills
- Architecture:distributed systems, microservices (Spring Cloud / Dubbo), API gateways, multi-tenant platforms, high availability and disaster recovery, domain decomposition, technology selection.
- Data and middleware:Kafka, RocketMQ, RabbitMQ, Elasticsearch, ClickHouse, Hive/Spark, Redis, MySQL, Oracle, MongoDB, Cassandra.
- Platform and operations:Docker, Kubernetes, Prometheus, CAT, Apollo, Sentinel, XXL-JOB, CI/CD, WebFlux.
- Management:team building and hiring (8–25 people), mentoring, planning with product, cross-team alignment, design and code review.
Experience
Independent — AI agent product | Family Learning Assistant
2024.11 – presentMoved from a Java backend to the AI agent stack, and designed and shipped an agent product for family use: a parent describes the goal in natural language and the agent understands intent, selects tools and orders the workflow to parse study material, organise vocabularies, generate practice and manage learning tasks. Source: github.com/sunxinshan/family-learning-assistant
- Agent architecture:a single agent with internal subgraph orchestration. The LLM only understands intent and selects tools; validation and routing are program-controlled, so domain rules never fall into free-form generation.
- Tooling:a unified tool registry owns definitions, permissions and implementations. Tools never touch the database directly — every business operation goes through the service layer, leaving room for on-demand loading and finer permissions.
- Human in the loop:interruption and resumption on LangGraph checkpointers; create, update and delete operations show the result first and execute only after confirmation.
- Engineering constraints:the backend is split into 11 domain modules; the service layer is the only entry point for data operations and cross-module repository access is forbidden; a dedicated user module owns authentication and data isolation.
- Stack:Python, FastAPI, LangGraph, React + TypeScript, MySQL, SSE, LangSmith; DeepSeek models.
Weizhi Technology| Senior System Architect
2022.03 – 2024.10Owned design and development of the core risk-control feature platform, production issue handling, and hiring and mentoring.
- Platform scope: configurable feature computation and variable accumulation for financial risk control, with a decision engine, case management and knowledge graph around it.
- Led the design of replay, backfill and high-availability solutions.
- Delivered graph feature design, ClickHouse integration and sub-tenant isolation.
- Wrote core code; responsible for hiring and developing the team.
Ximalaya| Engineering Manager
2021.07 – 2022.03Owned the marketing platform (PMS): tagging, placements, A/B testing, event centre, and user reach and delivery. Managed the PMS team (10+) and supported management and design of the LMS team (10+).
- Tagging:offline and real-time tag services for the education business. Offline supports T+1 audience pulls (incremental, reduced, full); real-time provides audience decision APIs.
- Offline and real-time tags share one definition and configuration layer that the engine resolves into different implementations, with Elasticsearch as the real-time store; real-time data comes from the event centre, offline from Hive.
- Event centre:real-time data processing and composite event handling. Input sources (Kafka, RocketMQ), output sinks (MQ/ES/HTTP) and processing logic are configurable; composite events such as "no claim 10 minutes after buying the trial camp" are supported.
- Placements:configuration-driven placement management for each client, covering banners and pop-ups, with audience segmentation deciding who sees what and when.
- A/B testing:a department-level experimentation system with flexible control and test groups.
- Reach and delivery:campaign delivery to target audiences through WeCom, official accounts, Qidian, Guangdiantong and Baidu.
- Defined long- and short-term platform goals with product, aligned stakeholders, led system decomposition and technology selection, controlled quality through design and code review, and owned stability monitoring and production incidents.
Ping An Financial Services| Engineering Manager
2018.09 – 2021.06Owned risk-related systems: LBS, external vendor gateway, risk-control middle platform and API gateway. Project owner for LBS, the gateway and the middle platform, and initiator of the API gateway project. Managed a team of 8+, served as tribe architect covering 50+ engineers, and later supported a 10+ person detection team and its technical design.
- LBS (real-time geospatial system):processes Ping An Bank user location data and provides data and product support to credit card, auto finance and next-generation banking across pre-loan, in-loan and post-loan stages. Deployed as separate data services, high-frequency API services, page services, admin and pre-production environments.
- Scale:80+ integrated external systems, 170+ interfaces, 4,000+ peak QPS, 200 million records a day, 1,000+ page users.
- External vendor gateway:the bank's single channel to external vendors, providing platform-level routing, rate limiting, circuit breaking, degradation and caching; 50+ vendors integrated, 1.5 million+ calls a day.
- Risk-control middle platform:integrated risk-control tribe products, guided users by business scenario, and extracted shared components (login, permissions, audit) to cut duplicate development; later evolved into a department-level platform.
- API gateway:a department-level open gateway supporting both Spring Cloud and Dubbo routing, with unified rate limiting and signature verification.
- Stack:LBS — Dubbo, CAT, Apollo, MySQL, ES, Kafka, RocketMQ, Redis, Sentinel; vendor gateway — Spring Cloud, MySQL/MongoDB, CAT, Apollo, Sentinel; middle platform — Spring Cloud, MySQL, Spring Cloud Gateway, Redis, RocketMQ; API gateway — Spring Cloud Gateway, Redis, RocketMQ, MySQL.
- Defined long- and short-term system goals with product, led decomposition, design and technology selection, controlled quality through review, and owned stability monitoring, vendor onboarding and production incidents.
Airwallex| Senior Software Architect
2017.06 – 2018.09Built channel systems for cross-border payments in China and Southeast Asia.
- FX distribution system:integrated upstream cross-border payment clients (foreign currency to CNY), distributed converted funds to domestic banks or third-party payment institutions, and settled into personal accounts.
- Stack:Spring Cloud, Cassandra, Redis, Docker.
- Led architecture design and delivery; mentored junior and senior engineers; resolved major technical issues.
Hewlett Packard Enterprise| Software Designer
2011.09 – 2017.06Developed and designed Service Manager, and built the containerised installation and configuration components.
- Service Manager:an ITIL implementation with the leading market share in ITIL, used by thousands of enterprise customers in Europe and North America. A highly customisable platform that lets customers tailor ITIL processes while shipping out-of-the-box functionality. Stack: Java, RabbitMQ, XSLT, RAD, Oracle.
- Docker installation and configuration:ITSMA (Service Manager, UCMDB, AM, Propel) required Kubernetes + Docker deployment; built install and configuration components embedded in the existing framework so administrators could change parameters during setup. Stack: React, Node.js, Java, Jenkins.
- Built a log analysis tool for diagnosing performance issues.
- Ran code and design reviews, delivered customer customisations, fixed production defects, supported the technical support team on complex issues, coordinated cross-team product integration, and ran internal knowledge sharing.
Education
Certifications & honours
- 2019 Best Architect of the Department, Ping An Bank
- 2020 Best Project Quality Award, Ping An Bank
- Alibaba Cloud ACP certified
- ITIL certified
- SAFe Scrum Master certified