Kun Lei

AI Full-Stack Product Engineer (Platforms, Productivity Tools, Rapid Prototypes)

Shanghai / Shenzhen / Suzhou · Currently employed; start date negotiable

Microsoft Software Engineer 2 delivering end-to-end engineering solutions across React product UX, backend integration, prompts and LLM evaluation.

Work Experience

2021.02 – Present

Software Engineer 2 (since 2022.09) · M365 Copilot Data Connectors (2026); MSN Finance & Community (2021–2025).

2020.05 – 2021.01

Software Engineer · React/TypeScript, Java/Spring marketing analytics; Jenkins, Docker and API tests.

2018.11 – 2020.04

Development Team Leader · Led five engineers on computer vision and Python/Flask model services.

2025–2026

Financial summary UX: Used React to build stock, index and currency price-movement summaries shipped on MSN Money product pages.
Agent evaluation data-labeling platform: Used React to build financial-summary fact-labeling and nine-dimension scoring review tools, creating human-reviewed golden datasets for agent quality evaluation and iteration.
Fact checking: Used LLM-as-Judge to verify financial AI claims against source data and expose traceable evidence for factual errors.
Earnings calls: Used audio/video API integration to deliver live earnings-call playback in the earnings hub.

Personal project

AI-assisted delivery: Use Codex + GPT6.1-Sol to implement and deploy each personal project in 3 hours.
Shared stack: Use React + TypeScript for product UX and Bun for service backends and shared gateway integration.
Nginx network tool: Use Nginx for domain routing and reverse proxies, using Git to selectively release and roll back application versions.
ClawNet: Use WebSocket to control remote coding agents from web/mobile, extending development beyond the workstation. Project link: https://clawnet.leikunx.com/
Friend-Connect: Use RAG to match group members by skills and needs with cited profiles, reducing discovery effort. Project link: https://friend-connect.leikunx.com/
App Market: Use LLM dialogue to clarify requirements and budgets in shared developer conversations, reducing handoff friction. Project link: https://app-market.leikunx.com/
Analytics: Use Azure Log Analytics for cross-site PV/UV and traffic trends, informing product iteration (owner login). Project link: https://analytics.leikunx.com/
Computer Monitor: Use Bun to collect per-second CPU/memory samples across computers, revealing resource bottlenecks. Project link: https://monitor.leikunx.com/
Cook: Use RAG over 372 recipes and 18 guides for source-linked cooking answers, reducing content-search effort. Project link: https://cook.leikunx.com/

2024–2025

Prompt templates: Used React to build a tool for six finance/sports scenarios with 100+ saved templates.
Agent tuning: Used React to deliver a prompt-tuning tool in one week for five teammates.
Backend prototype: Used Python to build live/batch financial queries in two days with a reusable core library.
Summary iteration: Used side-by-side evaluation to compare cross-market outputs as financial-summary tech lead.

2022–2023

Poll generation: Used an AIGC backend pipeline to generate polls for 10K+ trending articles daily.
Community UX: Used React to ship the AIGC poll experience reaching 100K+ consumption DAU.
AI writing: Used postMessage to connect an embedded writing assistant with the host post editor.
Operational visibility: Used real-time data queries to build dashboards, cutting latency from two days to five minutes.

Xiamen University · BEng, Software Engineering

2013.09 – 2018.07