Full-Stack & AI Engineer • Building since 2024

I build systems people actually ship.

Five production-grade platforms across full-stack web and applied AI. Real architecture. Real scale. See them for yourself →

View the work → Get in touch
01 / LabLemoLatex

LemoLatex

Real-time collaborative LaTeX editor

02 / Lablemo-ai

lemo-ai

Multi-agent reasoning platform

03 / MineSummary

Summary

Agent-native RAG study system

04 / LabPresenton

Presenton

AI presentation generator

05 / LabLemomateMAG

LemomateMAG

Enterprise ERP + collaboration suite

Scroll
0
Production systems shipped
0
Lines of code authored
0
Core technologies
0
In a research lab since
01 — Selected Work

Things I designed,
engineered & shipped.

LemomateMAG screenshot 1 LemomateMAG screenshot 2
Lab project
01

LemomateMAG

Enterprise collaboration + ERP suite (~48K LOC)

A unified workplace platform: instant messaging, project management, a full ERP (sales / purchasing / inventory / finance), HR, e-commerce, approval workflows and video meetings — 18 toggleable modules across 120 SQL migrations and 92 handlers.

  • Full ERP document flow: quote → order → delivery → invoice
  • Two-layer permission system: RBAC + data-scope, synced front/back
  • Multi-tenant isolation (company → department → employee)
  • Real-time WebSocket hub + LiveKit WebRTC video conferencing
GoGinReactRedux ToolkitTailwindLiveKitPostgreSQL
LemoLatex screenshot 1 LemoLatex screenshot 2
Lab project
02

LemoLatex

Real-time collaborative LaTeX editor — an Overleaf alternative

A full online LaTeX platform: server-side Docker compilation with a WASM fast-preview path, live multi-user editing, AI-assisted writing, and reference management. I built the compiler container pool, incremental compile cache, and the AI agent toolchain end-to-end.

  • Auto-scaling Docker compile pool with streaming logs over WebSocket
  • Incremental compilation via SHA-256 file hashing
  • CRDT real-time collaboration (Yjs + Monaco)
  • AES-256-GCM encryption of source at rest
GoFiberReactTypeScriptMonacoDockerPostgreSQLYjs
lemo-ai screenshot 1 lemo-ai screenshot 2 lemo-ai screenshot 3 lemo-ai screenshot 4
Lab project
03

lemo-ai

Multi-agent AI reasoning & orchestration platform

A three-tier reasoning engine (fast / standard / deepthink) with a multi-role "DeepThink" group-chat runtime, knowledge-base RAG, voice notes & podcast generation, speaker diarization, and a cross-session memory system backed by a Neo4j knowledge graph.

  • Multi-agent debate runtime — analyst, critic & synthesizer loops
  • Three-tier concurrency control with vLLM back-pressure
  • Document pipeline with LaTeX formula sprite-sheets
  • Podcast generation: outline → script → dual-voice TTS + FFmpeg mix
FastAPIVue 3Neo4jRedislitellmvLLMDocker
ai-review screenshot 1 ai-review screenshot 2 ai-review screenshot 3 ai-review screenshot 4
My own product
04

Summary / ai-review

Agent-native RAG study & revision system — my concept, my build

A self-initiated product: a three-tier microservice study assistant that turns any document into flashcards, mind-maps, quizzes and a human-AI co-created study plan. Driven by a ReAct agent over hybrid retrieval — not a simple RAG chatbot.

  • Hybrid retrieval: vector + BM25 → RRF fusion → cross-encoder rerank
  • Agent-native RAG with 12+ native tool-calling tools
  • Human-AI study-plan co-creation via interactive questionnaires
  • Dual-threshold context compression & multi-layer memory
Vue 3Spring BootFastAPIpgvectorLangChainPostgreSQL
Presenton screenshot 1 Presenton screenshot 2 Presenton screenshot 3
Lab project
05

Presenton

Local-first AI presentation generator

Generates full slide decks from a prompt or imported content (PDF, Word, web, CSV), with PPTX/PDF export and HTML+Tailwind custom templates. A unified multi-LLM client and an XML-level PowerPoint engine power it.

  • Unified multi-LLM client (OpenAI / Gemini / Claude / Ollama)
  • XML-level PPTX engine — rounded masks, shadows, advanced styling
  • Three-step pipeline: outline → layout match → content + async images
  • Auto-generated MCP server from the OpenAPI spec
FastAPINext.jspython-pptxSQLModelRedisChromaDB
02 — About

I'm a computer-science undergraduate who has spent the last two years doing the work most people only read about — shipping real, production-grade systems inside a research lab.

Since 2024 I've taken end-to-end ownership of platforms that span the full stack: Go and Python backends, React and Vue frontends, Docker microservices, and applied AI — RAG, multi-agent systems, and LLM orchestration. Most were built to a lab director's specification; Summary is my own product, from idea to deployment.

What I care about is the part that doesn't show up in a tutorial: container pools, incremental compile caches, hybrid retrieval pipelines, permission systems, encryption at rest. The unglamorous engineering that makes software trustworthy.

I work fast, I leverage AI as a force multiplier rather than a crutch, and I ship. If you need someone who can take a vague requirement and return a working, well-architected system — let's talk.

03 — Toolbox
04 — Contact

Have something worth building?

GitHub ↗ WeChat: startrek668