Enterprise AI surfaces
Open to work: AI Product Engineer / Frontend-heavy Fullstack Engineer
AI product builder / Frontend-heavy fullstack engineer
I build AI products that expose the work.
I am Larry Xue, also known as Yujian Xue. I am looking for AI product engineering work where I can turn rough workflows into usable agent products, enterprise interfaces, and public proof that a team can inspect.
- Enterprise surface
- CDN consoles, AI gateway UI, operations workflows
- Agent products
- 炼云小说工厂, YouChannel, agent workflow writing
- Open-source proof
- OpenClaw proxy patch, storage UI, build notes
Builder story
The through-line is product judgment under real constraints.
I learned reliability and complexity from cloud control panels. I learned AI product shape by building voice, memory, and workflow prototypes. I learned distribution by publishing open-source tools and writing the tradeoffs down.
AI-native engineering
I use AI as leverage, then keep the human judgment loop.
My workflow turns specs, prompt files, generated code, tests, and code review into repeatable delivery for admin-style product surfaces. prompt specs, test loops, review discipline, scaffold modernizationAgent product prototypes
I build the product loop before polishing the pitch.
Zota and Fluently explored session memory, context compaction, live voice, transcripts, learning records, and AI communication practice. Next.js, Supabase, AI SDK, Gemini Live, assistant-uiOpen-source proof
I leave work that can be inspected and reused.
Recent work includes Lianyun Novel Factory, an OpenClaw Feishu/Lark proxy fix, storage-console exploration, and public writing that records the tradeoffs. open source, upstream PRs, storage UI, public build notesWhat I would show first
Current work that maps to AI startups.
open-source May 2026 release
炼云小说工厂
An open-source AI web novel production workbench with agent workflows, prompt management, living story documents, batch jobs, and run tracing.
sanitized engineering case study
Enterprise AI Console Case Study
A sanitized reconstruction of CDN console, AI gateway, log search, async task, and internal operations work where dense configuration needs safe, inspectable UX.
upstream proxy fix and contribution fork
OpenClaw Contribution Fork
A focused OpenClaw contribution fork used to reproduce and patch a Feishu/Lark websocket proxy routing bug, then hand the proof back upstream.
knowledge workflow experiment
Obsidian LLM Wiki
A lightweight knowledge-base experiment around Obsidian-style notes and LLM-readable personal memory surfaces.
public product prototype
YouChannel
A YouTube-centered AI product experiment with a public web surface and companion service repo for video-aware workflows.
storage console exploration
MinIO Web UI
A Vue-based object storage UI exploration around bucket browsing and storage-console workflows.
Operating mode
How I build when the product is still unclear.
Start with the real workflow
A good AI product is not a demo prompt. It has users, state, retries, approvals, memory, logs, and a reason to exist inside daily work.
Expose the loop
Agents become usable when people can see what happened: inputs, tools, drafts, traces, failures, and where human judgment entered.
Ship small, then earn complexity
I prefer a working product slice with clear evidence over a large speculative architecture. The next abstraction should be earned by usage.
Research map
The questions I keep returning to.
Agent workflow architecture
Exploring how agent products should expose state, tools, approvals, memory, traces, failures, and user control.
AI communication practice
Testing how live voice, role play, feedback, transcripts, and memory can make language and work communication practice useful.
Enterprise AI interfaces
Designing AI and cloud product surfaces that keep complex configuration, logs, tasks, and review states understandable.
Large-model search visibility
Making personal sites and project pages easier for AI search systems to crawl, cite, summarize, and attribute.
Latest writing
Notes from the workbench.
记录炼云小说工厂 v0.1 的开源过程,以及我对 AI 长篇写作、agent 工作流和可观察生产系统的一些想法。 Related project: 炼云小说工厂.
Apr 29, 2026 More Agents Means More ManagementWhen one agent becomes many agents, the bottleneck moves from model capability to task, context, permission, state, and acceptance management. Related project: Enterprise AI Console Case Study.
Apr 23, 2026 Context Rot Is Two Different FailuresContext rot is not one disease. Long-context distraction and poisoned-context inertia need different product and engineering responses. Related project: Enterprise AI Console Case Study.