About
Python engineer with 7+ years of experience, including 5+ years on-site at WeChat Pay building test infrastructure for money-movement products. I designed the shared merchant fund test SDK, now reused by 23 business classes across 5 product areas, and was a core contributor to an api testability library, taking two core payment lines from hand-written scripts to reusable APIs on a unified SDK. Over the past two years I have brought LLMs and AI agents into test production with a staged spec-to-test pipeline, delivering new-product acceptance tests about 7× faster. English-major background; I work directly from English primary sources.
From English studies to engineering
I studied English at Chang’an University, then worked in overseas technical support at Tenda and Reolink, diagnosing networking issues for customers and channel partners worldwide in English. After years of troubleshooting other people’s systems I wanted to build reliable ones, so in 2018 I spent a full year studying Python, moved into backend development in 2019, and have worked on-site at WeChat Pay as a test engineer since 2021. Switching fields taught me to build knowledge from first principles; front-line support taught me to break a problem down before answering it; and my language background lets me read papers and primary sources directly and write design documents in both Chinese and English. One thread runs through it all: turning repeated business logic into maintainable, reusable engineering capability.
How I approach the work
I start by defining the problem and its boundaries, then design an implementation path that can be verified. I do not hide flaky failures behind reruns: I traced intermittent signature failures and found a hidden API defect, and I built a declarative retry-on-return-code decorator, promoted it to the shared framework layer, and used it to turn an async settlement window in a fund-allocation flow into self-healing standard retries instead of hand-written retry loops. When working with AI, I own requirements, design trade-offs, reviews, and acceptance while AI handles implementation and regression, and versioned documents and regression evidence keep every conclusion traceable.
Current interests
I am productizing what I learned on payments test platforms: a reusable HTTP API testing toolkit, an AI testing workflow, and agent tool contracts. In parallel I am studying LLM serving in depth: prefill and decode, KV cache, prefix caching, and load-testing methodology. Project cases describe delivered work, validation, and limits; notes keep the reasoning and the open questions.