---
title: "Hao-Ting Lee (李浩廷) · Software Engineer, SRE & AI Systems"
description: "Hao-Ting Lee (李浩廷, Justin) is a software engineer in Taoyuan, Taiwan who builds and operates reliable software and AI systems: lab monitoring, internal AI agent tools, local LLM inference and medical-imaging research."
image: https://dst-justin.com/img/og.jpg
url: https://dst-justin.com/
---

Skip to content

Software EngineerTaoyuan, Taiwan

Hi, I'm

# Hao-Ting Lee

李浩廷 · Call me Justin.

I turn _real-world needs_ into systems that keep running. My edge: a medical-imaging research background, hands-on hardware experience, and a habit of checking the numbers.

Reliable systems × AIResearch backgroundHardware & edgeReliable tools & systems

Read my story[Say hello](mailto:justin30303.jl@gmail.com)

HAO-TING LEESOFTWARE ENGINEER

KPITraction

Operations · LabMonitor

2,016

Max scheduled readings per day across 7 instruments

8

Manual recording rounds replaced every day

99.79%

Device-health snapshot marked OK over 24 hours

15,831

Successful runs of four scheduled jobs in August 2026

Engineering

2,091

Automated tests gating every LabMonitor deploy

6×

Lower upfront cost than the alternative equipment (NT$20k vs NT$120k)

8.2×

Lower p50 latency for a Relay helper rewritten in Rust (local benchmark)

33%

Smaller image than python:3.12-slim, app included (devops-lab)

Research & growth

7

Publications in journals and international conferences

3

ICMHI awards, including two first places

2

Where I presented research: Japan and the US

4+

Full-stack systems shipped as an intern

01The problem I solve

## Great demos are easy.Systems that keep running are not.

I started in a research lab, using deep learning to read brain CT and cell microscopy images. An internship at a biotech startup showed me the real gap: models and prototypes are everywhere, but things that run every day are rare. So that became my focus: turning lab needs into systems, wiring up cameras, sensors and GPUs, and finding out why when they break at midnight.

Justin · Taoyuan, autumn 2026

02My story so far

## From a research labto systems that run every day.

2020Research

### Starting in a research lab

Yuan Ze University · B.Sc. → M.Sc., Computer Science & Engineering

Researched medical imaging: hemorrhage localization, coronary calcification, circulating tumor cell segmentation. Ran Ubuntu lab servers and deployed LLMs on A100s, **the first time systems themselves caught my interest**.

2022First operations

### First taste of real operations

Taiwan Advanced Nanotech (TANBead) · MIS Intern

Helped manage internal systems and IT infrastructure, and handled whatever came up. **For the first time, textbook knowledge had to answer to real people.**

2023Building

### Turning needs into systems

Cancerfree Biotech · Software Engineer Intern

Research and product, side by side: 4+ full-stack systems with Vue.js and FastAPI, run as Docker services on GCP Compute Engine, watched by Uptime Kuma, YOLO on edge devices. **The research went on to Japan and Chicago.**

2025Operating

### Keeping systems running, every day

Cancerfree Biotech · Software Engineer

Grew internship prototypes into systems that run on their own, and own the moments they break. **I also build the internal tools colleagues use every day**: a company AI desktop assistant, a lab records app that replaced a paid SaaS, a Shopify back office, plus local AI inference and a PubMed RAG pipeline.

2026 →What's next

### Building systems teams rely on

Open to backend, internal tools, platform, AI engineering and SRE roles

Next is depth: **becoming an engineer who designs, deploys and operates reliable software and AI systems**, in Taiwan or abroad. I documented the path in my iThome Ironman series.

03What I bring

## Four strengths,one engineer.

The value I bring to a team. Each strength is backed by systems that have run in production, not just tutorials.

Full-stack systems

### Build

From requirements to a working product: APIs, databases, auth and the interfaces people actually use.

-   FastAPI · React · Next.js · Vue.js
-   PostgreSQL · Supabase · Neon · pgvector
-   Internal tools for lab, operations and e-commerce teams

Deploy · observe · recover

### Run

Keeping systems healthy after launch: containers, CI, monitoring, alerting and fixing what breaks.

-   Docker · GCP · GitHub Actions · Vercel
-   Prometheus · Loki · Grafana
-   ~12 alert conditions with confirm and re-notify

Hardware & edge

### Connect

Bringing software into the physical lab: cameras, sensors, edge boards and GPU servers.

-   Raspberry Pi 5 · Jetson Nano · DGX Spark
-   YOLOv8 and YOLO11 on edge devices
-   LLM and speech serving with vLLM and Ollama

Applied AI & imaging

### Research

A foundation in deep learning for medical images, now applied to agent tools and RAG, with the habit of measuring carefully.

-   PyTorch · OpenCV · CTC segmentation
-   Agent tools, RAG and MCP integration
-   7 publications, 3 ICMHI awards

04How I work

## Beyond the skills,this is how I work.

01 / 06

### Research first,  
then bring options

Ambiguity

When requirements are fuzzy, I don't wait around or start on a hunch. I **research the problem and possible directions myself**, lay out a few options, then align with whoever asked so we choose a path together.

Today that means turning a colleague's request into a UI or feature design, having Claude Code or Codex draft the plan, reviewing it, then shipping with tests, CI and a trial run by the person who asked.

02 / 06

### If it doesn't work,  
find another way

No sunk costs

For LabMonitor I first trained my own model to read instrument digits. The data was too uniform and it never got reliable. Instead of forcing it, I **switched to GPT-4o mini to read the digits** and got the system working first.

Version two now samples 7 instruments every 5 minutes, replacing 8 manual recording rounds a day.

03 / 06

### I like machines  
I can touch

Hardware

More than code that only lives in the cloud, I enjoy connecting things to the real world: **YOLO on Raspberry Pi and Jetson Nano, LLM and speech models on DGX Spark and Mac mini**, A100 lab servers at university.

Three sync loops on a Raspberry Pi 5 are supervised independently. One failing never takes down the others.

04 / 06

### Clear numbers,  
clear limits

Honesty

I care how a number was measured. Benchmarking Relay, I ran 3 warm-ups and 30 interleaved runs and **stated plainly that the results are local only**. In devops-lab I wrote down the scanner gap I couldn't solve instead of pretending I had.

If I didn't measure it, I don't call it a result.

05 / 06

### What the podium  
taught me about teams

Wind band conductor

In university I spent a year as the wind band's student conductor: planning rehearsals, choosing repertoire, working with instructors. **Getting dozens of people moving in one direction** takes planning and consensus, not orders. I bring that to work.

Handled our entry and preparation for the National Student Music Competition.

06 / 06

### Off the clock,  
still building

Curiosity

When I want to understand something, I build one. To manage my own dev accounts I wrote **Relay, a cross-platform CLI published on npm**. To learn the delivery chain I keep devops-lab: one small app taken through image signing, Kubernetes, Terraform and CI.

I also wrote a 30-day 2026 iThome Ironman series, Learning SRE for the AI Era.

05Case studies

## The strengths,in use.

### LabMonitor

Lab instrument monitoring

From a vendor-independent internship prototype to a second version with cameras, OCR, sensors and about 12 alert types, with no service outage while in operation. Materials cost about NT$20,000 versus an NT$120,000 alternative.

→ Build · Run · Connect

### CF OpenWorker

Internal AI agent platform

Colleagues needed AI that could work with confidential data. I led the company customization of an open-source desktop agent: company SSO, the internal model gateway, beta/LTS releases with auto-update, a web portal, and 7 agent tool families such as long-audio transcription and approval-gated sharing.

→ Build · Run · Research

### LabReport

Lab records · replaced Ragic

Moved the lab's records off Ragic, a paid SaaS, into an in-house app built around its own workflows: reagent lots, per-tube checkout and instrument records, behind company Azure AD. Four subscription seats are already cancelled.

→ Build

### Local AI inference

DGX Spark · Mac mini · vLLM

Set up and operate LLM, speech and video-generation models, with health checks and auto-recovery from GPU memory errors.

→ Run · Connect

### Prometheus VMS

Smart-glasses lab system

Worked on video management integration: multi-stream RTSP, live browser viewing, recording and clip extraction.

→ Build · Connect

### PubBrain

PubMed knowledge layer and RAG

Turned PubMed abstracts from 2016 to 2026 into a semantically searchable knowledge base on PostgreSQL and pgvector, with LangSmith tracing of the RAG flow.

→ Build · Research

### Eatable

Personal project · MCP server

A chat-driven meal logger with its own OAuth-protected MCP server: about 20 tools shared by the in-app agent and by Claude, plus a matching Agent Skill. I use it with Claude every day.

→ How I work: still building

### Relay

Personal project · npm

A cross-platform CLI for managing Claude Code credential snapshots, with locked writes and atomic file replacement.

→ How I work: still building

### devops-lab

Learning lab · open source

One FastAPI app carried through the delivery chain: a signed chiselled Ubuntu image, a hardened Kubernetes deployment that rejects unsigned images, Terraform-managed monitoring, and CI on kind. Runs on a laptop, not production.

[View on GitHub](https://github.com/darkstar1227/devops-lab)

→ Run · How I work: clear limits

06Talks & recognition

## Telling the worldabout the research.

I enjoy getting on stage and making something complex clear. My master's work on circulating tumor cells went to Japan and the US.

2022

ICMHI · Japan

### Locating brain hemorrhage

My first international talk: deep learning to locate hemorrhage in stroke patients.

Session Best Presenter

2023

ICMHI · Japan

### Detecting coronary calcification

First-author paper on detecting coronary artery calcification with deep learning.

1st Place · Student Essay

2025

ICMHI · Japan

### Segmenting CTCs over time

My master's research on stage: automated segmentation of cultured circulating tumor cells.

1st Place · Master's Presentation

2025

AACR Annual Meeting · Chicago

### Monitoring CTC tumoroids

First-author poster on an automated approach for real-time tumoroid monitoring.

2025

Diagnostics · journal

### Detecting demodex mites

Co-author of a peer-reviewed paper in Diagnostics 15(24).

[Read the paper](https://doi.org/10.3390/diagnostics15243204)

2025.07

Yuan Ze University · M.Sc. thesis

### CTC image segmentation

Cell segmentation, time-series and one-class anomaly detection. Advisor: Dr. Ting-Ying Chien.

07Life

## Outside of work,I'm happiest with people.

Music, snow and good company: the parts of me that don't fit on a resume.

Music

### From trombone to the podium

I played trombone in the Yuan Ze University wind band, then spent 2021–2022 as its student conductor: leading rehearsals and performances, choosing repertoire, and preparing us for the National Student Music Competition.

Snowboarding

### Ten days in Hokkaido, every winter

Snowboarding is my favourite way to switch off. For the past three winters I have spent about ten days riding in Hokkaido.

Group activities

### Better together

I love joining group activities. It's the same reason I loved the band: lots of people, one shared goal.

08Roadmap

## Still exploring,with a clear direction.

I'm in no rush to pin myself to one job title. Over the next three years I want to become an engineer who can **design, deploy and operate reliable systems**, finding my specialty through hands-on work.

-   **Most interested**Reliable backend and AI systems, internal tools, infrastructure and operations
-   **Also love**Hardware, compute infrastructure, device deployment, system integration
-   **Still pursuing**Self-directed research in biomedical and applied AI
-   **Location**Taiwan first; open to international roles and relocation
-   **Languages**Mandarin (native) · English for work (IELTS 6.0)

### Daily drivers

-   Python
-   FastAPI
-   Docker
-   PostgreSQL
-   Linux / Ubuntu
-   Claude Code
-   Codex

### Deploy & observe

-   Docker Compose
-   GCP
-   GitHub Actions
-   Vercel
-   Supabase
-   Cloudflare
-   Prometheus
-   Grafana
-   Loki
-   LangSmith

### AI & hardware

-   PyTorch
-   OpenCV
-   YOLO
-   vLLM
-   Ollama
-   CUDA
-   DGX Spark
-   Jetson Nano
-   Raspberry Pi

### Frontend & other languages

-   Vue.js
-   React
-   Next.js
-   C++
-   Rust (beginner)

TypeScript work is AI-assisted.

2026 iThome Ironman · AI Engineering

### Learning SRE for the AI Era

My contest series, from an SRE lab to production AI reliability: FastAPI, Prometheus, Grafana, Loki, OpenTelemetry and LLM tracing, applied to SLOs, error budgets, failure injection and incidents.

Build → Trace → Break → Measure → Evaluate → Recover → Improve

#### Selected articles (in Traditional Chinese)

1.  [Day 04What Is SRE? Reliability Is Not Zero FailureSRE manages service reliability with software engineering, and the job goes far beyond monitoring and firefighting.](https://ithelp.ithome.com.tw/users/20163611/ironman/9343)
2.  [Day 07Technical Success ≠ Semantic SuccessHTTP 200 and an all-green trace prove the pipeline ran, not that the user got a usable, correct and safe answer.](https://ithelp.ithome.com.tw/users/20163611/ironman/9343)
3.  [Day 14SLI, SLO, SLA: Measure the Promise FirstAn SLI is a measurement, an SLO a team target, an SLA an external commitment; mixing them makes dashboards and contracts disagree.](https://ithelp.ithome.com.tw/users/20163611/ironman/9343)
4.  [Day 16Why Average Latency LiesAverages hide a few very slow requests; interactive services need the distribution and P50, P95 or P99.](https://ithelp.ithome.com.tw/users/20163611/ironman/9343)
5.  [Day 20Error Budget and Burn RateAn error budget is the risk an SLO accepts, not a quota for failure; burn rate says whether you will spend it too early.](https://ithelp.ithome.com.tw/users/20163611/ironman/9343)
6.  [Day 26AI Alerting: One Hallucination Shouldn't Wake EveryoneRoute AI alerts by impact and actionability: outages page on-call, while single quality samples go to evaluation or review.](https://ithelp.ithome.com.tw/users/20163611/ironman/9343)

[Read the series on iThome →](https://ithelp.ithome.com.tw/users/20163611/ironman/9343)[Lab code on GitHub →](https://github.com/darkstar1227/learning-sre-for-ai-era)

09Let's talk

## Let's build together.

Hiring for backend, internal tools, platform, AI engineering or SRE? Looking for a research collaborator? I'd love to hear from you.

[justin30303.jl@gmail.com](mailto:justin30303.jl@gmail.com)[LinkedIn](https://www.linkedin.com/in/justin-lee-dst1227)[GitHub](https://github.com/darkstar1227)

© 2026 Hao-Ting Lee 李浩廷Built with the Prism design system

```json
{
  "@context": "https://schema.org",
  "@graph": [
    {
      "@type": "WebSite",
      "@id": "https://dst-justin.com/#website",
      "url": "https://dst-justin.com/",
      "name": "Hao-Ting Lee",
      "inLanguage": [
        "en",
        "zh-Hant"
      ],
      "publisher": {
        "@id": "https://dst-justin.com/#person"
      }
    },
    {
      "@type": "ProfilePage",
      "@id": "https://dst-justin.com/#profile",
      "url": "https://dst-justin.com/",
      "name": "Hao-Ting Lee (李浩廷) · Software Engineer",
      "isPartOf": {
        "@id": "https://dst-justin.com/#website"
      },
      "mainEntity": {
        "@id": "https://dst-justin.com/#person"
      },
      "dateModified": "2026-11-07"
    },
    {
      "@type": "Person",
      "@id": "https://dst-justin.com/#person",
      "name": "Hao-Ting Lee",
      "alternateName": [
        "李浩廷",
        "Justin Lee",
        "Justin"
      ],
      "givenName": "Hao-Ting",
      "familyName": "Lee",
      "url": "https://dst-justin.com/",
      "image": "https://dst-justin.com/img/portrait-960.webp",
      "email": "mailto:justin30303.jl@gmail.com",
      "jobTitle": "Software Engineer",
      "description": "Software engineer who builds and operates reliable software and AI systems, with a background in deep learning for medical imaging.",
      "address": {
        "@type": "PostalAddress",
        "addressLocality": "Taoyuan",
        "addressCountry": "TW"
      },
      "worksFor": {
        "@type": "Organization",
        "name": "Cancerfree Biotech Ltd",
        "alternateName": "精拓生技股份有限公司",
        "url": "https://cancerfree.io"
      },
      "alumniOf": {
        "@type": "CollegeOrUniversity",
        "name": "Yuan Ze University",
        "alternateName": "元智大學"
      },
      "hasCredential": [
        {
          "@type": "EducationalOccupationalCredential",
          "credentialCategory": "degree",
          "name": "M.Sc. in Computer Science and Engineering",
          "recognizedBy": {
            "@type": "CollegeOrUniversity",
            "name": "Yuan Ze University"
          }
        },
        {
          "@type": "EducationalOccupationalCredential",
          "credentialCategory": "degree",
          "name": "B.Sc. in Computer Science and Engineering",
          "recognizedBy": {
            "@type": "CollegeOrUniversity",
            "name": "Yuan Ze University"
          }
        }
      ],
      "knowsAbout": [
        "Site Reliability Engineering",
        "Infrastructure",
        "Full-stack development",
        "FastAPI",
        "React",
        "PostgreSQL",
        "Docker",
        "Prometheus",
        "Grafana",
        "LLM inference",
        "vLLM",
        "Retrieval-augmented generation",
        "Model Context Protocol",
        "Edge AI",
        "Medical imaging",
        "Deep learning"
      ],
      "knowsLanguage": [
        "zh-Hant",
        "en"
      ],
      "award": [
        "1st Place, Master's Presentation, ICMHI 2025",
        "1st Place, Student Essay Competition, ICMHI 2023",
        "Session Best Presenter, ICMHI 2022"
      ],
      "sameAs": [
        "https://github.com/darkstar1227",
        "https://www.linkedin.com/in/justin-lee-dst1227",
        "https://ithelp.ithome.com.tw/users/20163611/ironman"
      ]
    },
    {
      "@type": "CreativeWorkSeries",
      "name": "Learning SRE for the AI Era",
      "url": "https://ithelp.ithome.com.tw/users/20163611/ironman/9343",
      "author": {
        "@id": "https://dst-justin.com/#person"
      },
      "inLanguage": "zh-Hant",
      "description": "2026 iThome Ironman series (AI Engineering): from an SRE lab to production AI reliability."
    },
    {
      "@type": "ScholarlyArticle",
      "headline": "Demodicosis Mite Detection in Eyes with Blepharitis and Meibomian Gland Dysfunction Based on Deep Learning Model",
      "isPartOf": {
        "@type": "Periodical",
        "name": "Diagnostics"
      },
      "datePublished": "2025-12",
      "sameAs": "https://doi.org/10.3390/diagnostics15243204",
      "author": {
        "@id": "https://dst-justin.com/#person"
      }
    }
  ]
}
```
