Tiancheng Lu

Tiancheng Lu

AI Application Full-Stack Engineer · Frontend-focused

Frontend-led full-stack engineer, 3 years: product flow design, frontend implementation, model & workflow integration, admin systems, licensing, and shipping to production.

experience
3 yrs
experience
lingma-proxy on GitHub
38★
lingma-proxy on GitHub
engineering blog posts
13
engineering blog posts
View projects Contact me Engineering blog →

Featured Projects

Every project links to evidence: source code, live products, or engineering write-ups. Private projects are presented as sanitized case studies — each with system boundary, key decisions and verification.

lingma-proxy · Lingma-compatible proxy / Feishu Agent

Open source

Solo developer

Adapts Tongyi Lingma / Qoder local capabilities into standard OpenAI / Anthropic protocols so any agent client can plug in.

  • OpenAI / Anthropic / Responses protocol adaptation: model listing, streaming, parameter pass-through, error-format compatibility — works with Claude Code, Codex CLI, Cline and more
  • Dual backends (Remote API & IPC): WebSocket / Named Pipe probing, model fallback, multi-runtime compatibility; full Tools / Function Calling support
  • Feishu Agent branch: CardKit streaming replies, Skills / MCP tool extensions, SQLite / FTS5 tool memory, scheduled tasks and safe file access
GoWailsVue3WebSocketNamed PipeSQLite/FTS5MCP
GitHub · 38★ → Architecture write-up (CN) →

Outsider Studio · AI short-video production

Private · sanitized case

Solo developer

For POV short-video creators: an AI workbench from topic, script, storyboard, image generation and TTS to Jianying export.

  • Core pipeline: script generation, storyboard splitting, reference management, image gen, voice synthesis, video preview and Jianying export; multi-provider routing, fallback and config management
  • Commercialization loop: License + Admin console + distribution invite codes & quotas + soft-OTA release pipeline — desktop to cloud, end to end
  • Long-running generation UX: task state machine, failure retry, batch cancel, resume-from-checkpoint and history
Tauri 2React 19RustCloudflare Workers/D1/R2HonoFFmpeg
Product site → Read the case study →

Ecom Image · E-commerce AI asset workbench

Private · sanitized case

Solo developer

Full pipeline for e-commerce assets: hot-product collection as reference images, multi-type generation with human review before delivery.

  • Built-in crawl scheduler (APScheduler in the unified API): best-seller / store + keyword collection on major platforms, business-friendly cron config, two-layer idempotent dedup
  • Separate stores and tags for hot-product vs reference images with explicit promotion; references + locale/platform style profiles + versioned templates assemble the prompts
  • Full lineage: Project → MarketVersion → Job → DeliveryAsset, with QA rules and human review before delivery
Next.jsFastAPIPostgreSQLMinIOAPScheduler
Read the case study →

Job-opportunity screening Agent with human-in-the-loop

Private · sanitized case

Solo developer

Rule + LLM dual evaluation for job opportunities — real actions never fire from the model, only from human approval.

  • Full chain: source adapters → normalization & dedup → candidate profile → rule + LLM evaluation → human review queue
  • Analysis-only by default: thresholds, daily caps, risk cooldowns, queue limits and explicit confirmation as pre-action safety gates
  • Auditable end to end: execution batches, system events, review records and exclusion states
PythonLocal LLMSQLite
Read the case study →

Client Delivery

Two client projects running in parallel, built to production standards and heading to real deployment.

Foreign-trade digital-workforce platform

Architecture & full-stack delivery · production rollout in progress

A self-hosted digital-employee workspace for a foreign-trade enterprise: three AI digital employees cover lead qualification, outreach drafts, customer replies and quote/order documents across the full business path; every AI draft passes a human review gate before it reaches a customer. Dify serves only as the workflow and model engine — customers, inquiries, knowledge base, channels, permissions and audit all live in the self-built business core.

FastAPI · Next.js · Dify · PostgreSQL · MinIO

PDD / Douyin OMS

Architecture & delivery · production rollout in progress

A privately deployed order-management system for a merchant: PDD/Douyin order ingestion, review with inventory reservation, e-waybills, local printing, pickup write-back and inventory-ledger sync across the full fulfillment chain; designed for 100k–500k orders per day, built on an open-source ERP fork trimmed into a modular single-merchant architecture with elastically scaled integration workers.

Java 17 · Spring Boot · Vue3 · MySQL/Redis · Docker

Experience

2025.05 – 2026.07

AsiaInfo Technologies (Guangzhou) Frontend Engineer

  • Frontend development, API integration and launch support for China Mobile business systems: marketing campaigns, membership benefits, family plans, online contract termination, domestic-stack migration
  • Improved campaign-page loading, section switching, polling refresh and fallback logic — fewer white screens, jank and redundant renders on mobile
  • Built internal AI-agent productivity tools (requirements-doc generator, self-test report generator), cutting repetitive documentation work
2024.03 – 2025.05

Shenzhen Yunzhan Information Technology Frontend Engineer

  • Sole frontend engineer; owned the frontend architecture of an AIGC assistant platform: intelligent Q&A, document parsing, image generation, AI video modules
  • Integrated ChatGLM4, OpenAI, Azure OpenAI, DeepSeek, Midjourney, Kling and other generative models into web experiences
  • Led Dify / RAGFlow secondary development and business adaptation; Node.js + React admin system; n8n / Coze automation workflows; Docker packaging and server maintenance

Skills

Web fundamentals
HTML · CSS · JavaScript / ES6+ · TypeScript
Frontend frameworks
React 19 · Vue 2/3 · Vite · Next.js · React Router · uniapp
State & data
Zustand · TanStack Query · TanStack Virtual
UI & interaction
Tailwind CSS 4 · Radix UI / shadcn · Ant Design · motion
Desktop
Tauri 2 · Wails · Rust integration · desktop packaging · soft-OTA updates
AI application engineering
LLM API integration · multi-model routing · OpenAI / Anthropic-compatible APIs · Dify · RAGFlow · n8n · Coze · prompt engineering
Backend & ops
Node.js · Hono · Docker · Cloudflare Pages / Workers / D1 / R2 · Git · CI/CD

Selected Writing

Rebuilding WeChat Jump in two days: Unity WebGL launch + global leaderboard (CN) →

AI closed-loop debugging, a zero-dependency Node leaderboard, and bugs only real devices reveal.

Production LLM gateway: failover, semantic caching and cost control (CN) →

Engineering practices for multi-channel resilience and cost governance.

Deploying NewAPI as an AI gateway on Tencent Cloud (CN) →

From Docker Compose to HTTPS and channel configuration.

Tauri vs Electron for personal desktop tools (CN) →

A comparison grounded in real desktop-product delivery.

Contact

Resume and further project material available on request via email. Public timeline on the blog; source on GitHub.

GitHub Blog