ProfilePortrait
蔡承紘 · Tsai Cheng-Hung
My journey, left → right.
$ whoami▌
TSAI
CHENG-HUNG
蔡承紘 — AI Application · Full-Stack Engineer
I build AI that speaks human — RAG systems, autonomous agents, and the interfaces that make them effortless.
BOOT // SCATTEREDASM_000%
SYS/01
About
A medical-informatics graduate who fell for frontend craft.
AI Application & Full-Stack Engineer — building reliable AI systems and shipping real-world products.
> Hi, I'm Tsai Cheng-Hung!
I'm a Master's graduate from the AI track of Medical Informatics at Kaohsiung Medical University, specializing in RAG systems, agentic workflows, and the frontend interfaces that bring them to life.
With a focus on moving AI from prototype to production, I've built and deployed agentic workflows with LangGraph and RAG-driven Q&A pipelines — from a complete air-pollution thesis system to multi-agent consultation platforms. My strength lies in bridging large language models and robust full-stack systems: integrating vector databases, prompt and persona control, and guardrails that keep model output reliable and user-friendly.
SYS/02
Career Slices
A horizontal look back at every role I've worn — from hospital floors to AI engineering.
Profile
UndergraduateHealthcare Administration
I-Shou University
Where my healthcare foundation began.
InternshipHealthcare Quality Intern
National Defense Medical Center
Exposure to hospital quality & process.
HospitalPreventive Medicine Admin
E-Da Hospital
Records, tracking & health-check reports.
Master'sMaster's Researcher
Medical Informatics · KMU
Built & deployed a RAG thesis system.
TeachingAssistant TA
Taiwan AI Course Alliance
Guided ~100 students in generative AI.
.png&w=3840&q=75)
InternshipRAG Engineer (Intern)
學思行數位行銷
Agent flows, RAG & full-stack product.
CurrentAI Application Engineer
中華工程 / BES
Production RAG systems, agents & automation.
SCROLL ↓ — TRACK SLIDES →TRK_000%
SYS/03
Experience
1–2 years across AI engineering, full-stack, and teaching.
2026.04 — PRESENT● LIVE
AI Application Engineer
BES / 中華工程
Building production RAG systems, autonomous agents, and automation pipelines.
2025.08 — 2025.10
AI Engineer (Intern)
學思行數位行銷
Full-stack on a small product, frontend lead on a large one. Agent flows (LangGraph/LangChain), Flask & FastAPI, RAG, streaming responses; React/Next.js, Zustand, Prisma + PostgreSQL.
2025.03 — 2025.06
Teaching Assistant — Generative AI
Kaohsiung Medical University
Supported ~100 students in model application and coding; designed AI problem-solving tasks.
SYS/04
Toolkit
From embeddings to interfaces.
MOD.01AI / ML
PyTorchHugging FaceRAGLangGraphAgentic WorkflowPrompt EngineeringLoRA / unslothEmbeddings
MOD.02Backend
FastAPIFlaskLangChainOllamaChromaDBREST API
MOD.03Frontend
ReactNext.jsTypeScriptTailwindViteZustandRWD
MOD.04Data / Tools
PostgreSQLMySQLPrismaPineconeGitDocker
SYS/05
Selected Work
Systems shipped, not just demoed.
01↗02↗03↗04↗
Construction Knowledge RAG & Steel Procurement Meeting Automation
LangGraph / FastAPI / Hybrid RAG / ChromaDB / BM25 / RRF / Gemini Grounded Search / OpenAI Embeddings / SQLAlchemy / python-docx / Next.js / TypeScript
A production-oriented construction knowledge assistant and steel procurement meeting generator built with LangGraph, hybrid RAG, Gemini grounded search, structured extraction, and DOCX automation.
Eastern Mysticism Platform
LangGraph / GPT-4o / Quart / Redis / PostgreSQL / Pinecone / Next.js 15 / Three.js
Multi-agent consultation platform with SSE streaming and a custom Zi-Wei chart engine.
Multi-modal RAG Knowledge Base Platform
LightRAG / RAG / Ollama / ChromaDB / FastAPI / Next.js 16 / MinerU / Docker Compose
A production-grade, fully containerized multi-modal RAG platform with real-time SSE streaming, multi-turn conversation compaction, session-scoped retrieval, and an admin console — all running locally via Docker Compose.
Ziwei Doushu AI Platform — From a Take-Home Test to a Full Multi-Agent Product
LangGraph / Google Gemini / iztro / ChromaDB RAG / FastAPI / PostgreSQL / Next.js 14 / Docker
An account-based AI fortune-telling platform pairing a deterministic Ziwei Doushu chart engine (iztro) with a genuine parallel multi-agent LangGraph pipeline, RAG knowledge base, and a full subscription/token system — deployed on Docker + Render.
SYS/06
Publication
AIET 2026 · Zagreb, Croatia
ACCEPTED // ORAL