Smart MFG · AI Platform Engineer

I build AI applications across the full stack. Model accuracy is just the start — what matters is running reliably in production.

At AUO, I bring computer vision and LLM applications into manufacturing, handling everything from models, backend services, and data to deployment and ongoing operations.

These are the projects I build after hours, following the same principle and solving problems I care about.

Experience

What I've done

01 Smart Manufacturing Engineer 2021.06 — present
02 Cross-functional Team Lead, AI Vision Alongside the role above
03 Instructor, AUO University 2022 — present
04 Research Assistant 2018.06 — 2020.07 · part-time

Smart Manufacturing Engineer

AUO · L8B Houli Fab

2021.06 — present

AI vision platform: Led the design and build of a no-code AI platform covering data governance, labelling, training and validation, versioning and remote deployment. Used RabbitMQ for asynchronous tasks and NGINX for load balancing to support concurrent workloads and multi-user collaboration.
Deploying AI in manufacturing: Defect localisation and severity grading, electrical signal analysis, virtual metrology, SPC, automated repair and R2R control. Perspective transform, homography and image stitching for correction and automated measurement, plus real-time RTSP inference across multiple cameras.
Enterprise AI agent: Integrated vector databases, RAG, an LLM gateway, MCP and Azure Databricks, with responsibility for long-conversation context management, session-memory compaction and automatic summarisation. Built a natural-language search assistant using Elasticsearch and RAG to retrieve platform and project information.
Fab-wide management and operations: Built the metrics for the AI transformation programme, plus project tracking and server-health monitoring — scheduling, database redundancy and automatic alerting. Consolidated the web services and APIs from every department behind one reverse proxy.

Computer Vision
FastAPI
Python
MongoDB
MySQL
Redis
RabbitMQ
NGINX
RAG
MCP
Azure Databricks

Cross-functional Team Lead, AI Vision

AUO

Alongside the role above

Aligning technology across fabs: Led technical integration and functional alignment across fabs and departments, bringing state-of-the-art methods into manufacturing operations.
Shared tooling and SOPs: Supported colleagues building vision projects, wrote the SOPs and built shared tooling so no project had to reinvent the wheel.

Technical leadership
Cross-team collaboration
SOP

Instructor, AUO University

AUO

2022 — present

Teaching: Delivered two rounds of AI vision training for 21 departments and supported technology deployment across 16 sites. Served as a mentor in the AUO A+ internship programme for three consecutive years.
Speaking and hosting: Guided tours at the smart manufacturing exhibition and hosted technical visits from the WEF, Ford and BHTC. Spoke at the AUO smart manufacturing forum on getting AI into production and keeping fabs aligned.

Training
Mentor
Technical outreach

Research Assistant

Feng Chia University · Intelligent Transportation & Logistics Innovation Center

2018.06 — 2020.07 · part-time

The full cycle of commissioned research: One project for the Institute of Transportation, two for the Ministry of Science and Technology. Each had a client and a deliverable, so it ran the whole loop — agreeing model choice, data collection and schema with the vendor, then validating against field data, writing it up, and proposing what the next phase should fix.
Where the computer vision started: Traffic-incident detection: the logic and the pipeline, tuning lane occupancy, turning volume and speed. For mixed food products, object detection tagged the items, then the model was transfer-learned into a triplet network to extract features — a fine-grained comparison between the listing photo and the physical product that flags the ones that do not match. Both became journal papers.
Unstructured data analysis: Vulnerability and resilience in perishable food supply chains: turned papers, news and industry reports into structured data, mined association rules with SAS, and used k-fold cross-validation to narrow it down to the factors that mattered.

Object Detection
Transfer Learning
Triplet Network
SAS
Association Rules

Credentials

Achievements & Education

Achievements at AUO

AUO · 2021.06 — present

Publicly verifiable

Patent
Inspection system and inspection method
- Taiwan TWI845420B · granted
- US US12586160B2 · granted
- China CN118015400A · application published
Competition
AI CUP STAS Pathology Image Segmentation II
- 2022 · team won the Program Office Award

Contents withheld

Trade secret
Defect vision inspection and control
- Real-time inspection driving a PLC cycle-stop
Trade secret
Live equipment-condition vision control
- Defect detection driving RPA-based machine repair

The trade secrets stay closed under company policy — listed here by subject only.

Education

2024.09 —
National Chung Hsing University

Part-time master’s program in Computer Science
- Certification
  Microsoft Azure
  - AZ-900 · AI-900 · DP-900
2018 — 2020
Feng Chia University
GPA 4.0
Master’s degree in Transportation and Logistics
- Paper
  AI-based quality risk management in omnichannel operations: A potential solution to O2O food dissimilarity
  - International Journal of Logistics Management
- Paper
  An object-detection logic for traffic incidents, with empirical validation
  - Journal of the Chinese Institute of Transportation
- Competition
  CIIE Master’s Thesis Competition
  - 2020 · Award of Excellence, Service Systems & Technology Management
- Certification
  
  SOLE — Demonstrated Logistician
- Honour
  
  Phi Tau Phi Scholastic Honor Society · member
2014 — 2018

Feng Chia University
GPA 3.8
Bachelor’s degree in Transportation and Logistics

Capabilities

What I can do

Given a problem I weigh the whole thing — where the data comes from, what the model can do, the API, the architecture, the interface, where it runs and who keeps it running — rather than fixing on one model or one framework.

Across every layer

These capabilities span every layer. They shape how the layers work together and whether the system remains reliable.

AI Solution Architecture

Evaluating technologies and architecture based on requirements, then developing prototypes into systems that teams can deploy, operate and maintain.

Computer Vision
LLM
RAG
AI Agent
Embedding
Search
Workflow Automation

Architecture Optimization

Refactoring and optimising existing systems for performance, scalability, availability and maintainability.

Service Routing
Cache Strategy
Reverse Proxy
DB Optimization
Asynchronous Processing

Cross-Team Project Leadership

Aligning technical requirements across fabs and departments, establishing SOPs and delivering training to help other teams adopt the technology.

Technical Alignment
SOP
Shared Tooling
Training
Mentor

Application layer

Frontend & Application

Turning a model into something a person can use, not just an API endpoint.

Web Application
Responsive UI
PWA
Cross-Device Apps

Service layer

Backend & API

Integrating enterprise systems and services.

RESTful API
Microservice
Authentication
Async Processing

Distributed & Async

Designing services to handle concurrent workloads and process time-consuming tasks asynchronously.

Message Queue
Background Task
Concurrency
Task Scheduling

Model layer

Computer Vision & Industrial AI

Designing inference around what the process actually needs, not around a number on a held-out set.

Classification
Object Detection
Instance Segmentation
Anomaly Detection
OCR
DIP

Generative AI & Retrieval

Retrieval and generation over a company’s own knowledge.

Hybrid Retrieval
Reranking
Vector Search
Local LLM
LLM Gateway
Multi-Model

AI Agent & Automation

Taking an LLM past answering questions into operating real systems and workflows.

Tool Calling
MCP
Workflow Automation
Multi-Agent

Data layer

Data & Search

Large-scale data storage, caching and retrieval.

Relational / NoSQL
Cache
Full-Text Search
Vector Search
Embedding Retrieval

Side Projects

Services I run myself

One sign-in and permission model across all of them. Some need approval first — the card takes you to the request form.

ETAK Tactical Map
Invite only

A live shared tactical map for airsoft — everyone marks the same board, with rooms to keep games apart.

No public requests — I send the invite.
Dart Win
Approval required

Multiplayer darts scoring. Open it on a phone and play — no account needed.

Scan the QR code at the board and play. Ask for access only if you want the history kept.
AI Image
Invite only

An AI imaging platform — data and training in the cloud, all the way out to the phone.

No public requests — I send the invite.
Gopheru Trip Planner
Approval required

Itineraries and a travel handbook my partner and I keep in one place, together.
Edwin Room Pro
Approval required

A perspective-correction tool — straightens a photo shot from the wrong angle.

Static Pages

Small projects for everyday use

Quick references for everyday use.Sign in and the links show up here.

BEYBLADE X rankings Eight tops compared on stamina, defence and attack. Sign in for the link
Networking final revision 39 questions and the notes that matter, built to drill on a phone. Sign in for the link
Airsoft GBB pistol comparison Twelve gas pistols, reliability and accuracy set against price. Sign in for the link

The badge in each card's top corner

Approval required: Sign in with Google to continue. If you already have access, you can enter the service directly. Otherwise, submit an access request for my approval. Signing in does not automatically grant access.
Invite only: No public requests — I grant these myself. The card shows you how to reach me.