Data x backend x applied AI
Min-Huan Tsai portrait
Personal portraitSan Francisco

Min-Huan Tsai

I like systems work where performance, correctness, and actual deployment matter: pipeline latency, model cost, search quality, and user-facing reliability.

High-leverage engineering over surface-level demos.

At a glance

3

Internship experiences

4

International publications

3

First-author publications

1st Place

Hackathon finish

What I build

I build data products and software that hold up in production.

NCKU EE-CSIE master's student with internship experience at TSMC, GoFreight, and QuickClick. My work spans data engineering, backend systems, and applied AI, with four international publications, three internships, and a TSMC hackathon win.

Experience

Data Engineer Intern, RD

GoFreight

Sep 2025 - Mar 2026

Built production dbt models and analytics layers on Snowflake, improving both pipeline reliability and downstream decision speed.

Architected and deployed 10+ dbt models for downstream BI and reporting.
Optimized 3 core fact models and reduced build time by 60%.
Designed 4 Metabase dashboards that turned operational data into usable business views.

Software Engineer Intern, IT

TSMC

Jul 2025 - Aug 2025

Worked on document intelligence and search workflows, combining Java backend systems with LLaMA-based processing.

Developed an auto-summarization, tagging, and org-ID pipeline for 100K documents.
Designed a Trie-based matcher that ran 70% faster than the prior regex approach.
Cut LLaMA token usage by 95% through pre-extraction and filtering logic.

Software Engineer Intern, RD

QuickClick

Jun 2022 - Jan 2023

Shipped backend product features in a user-facing fintech environment with production traffic and strict correctness expectations.

Built a cashflow verification module serving 10,000+ users.
Delivered rebate and coupon modules with Node.js for large-scale user operations.

Publications

From Thought to Action: An Interactive Platform for Inspecting Strategic Reasoning in LLMs

2026 AAMAS Workshop Paper

3rd author

SAMPLE: A Spatial/Channel-wise Attention GCN with MLP and Periodic Linear Encoding for Land Boundary Demarcation System

2026 WWW Demo Paper

1st author

GRIDS: A Geospatial AI System for Land Boundary Demarcation

2025 ICDM Workshop Paper

1st author

MT-redis

Linux Foundation Open Source Summit 2025

3rd author

Projects

First-author · 2025 IEEE ICDM Demo Paper

GRIDS

Full-stack AI system for land boundary demarcation built with the Tainan City Government. Reduced processing time by 70% while holding prediction error within 10%.

Python, React, PyTorch, AWS, MySQL

1st Place · TSMC IT CareerHack 2025

Vision AI Assistant

Built an agentic vision assistant across 10+ project types, reaching 88% accuracy and reducing inference time by 50-80%.

Python Flask, React, Gemini, LLaMA, SAM2, DINO, Diffusion, GCP

Co-author · Linux Foundation Open Source Summit 2025

mt-redis

Benchmarked multi-threaded Redis variants and found throughput gains from QSBR and CPU affinity tuning under 8-thread workloads.

C, RCU, benchmarking, Linux perf

1st Place NCCU MIS 2024 · 3rd Place International ICT 2023

ImagicNation LLM TextBook

Built backend infrastructure for a generative textbook platform using GPT-4 and DALL·E 3 to create child-friendly story content.

Next.js, Node.js, Docker, MySQL, AWS

Core stack

Data Platforms

PythonSQLSnowflakedbtDagsterMetabaseELT pipelines

Backend Systems

JavaSpring BootNode.jsTypeScriptMySQLMongoDBFirebase

Applied AI

LLM pipelinesPyTorchComputer VisionPromptingRAG-style workflowsVertex AI

Infra

DockerKubernetesAWSGCPLinuxTesting

Education

NCKU EE-CSIE

Master of Electrical Engineering-CSIE, GPA 4.25/4.3, expected June 2026.

NCCU MIS

Bachelor of Management Information Systems, GPA 3.7/4.3.

Extra signal

Linux Kernel Internals by Jserv (A+), Algorithms (A+), Data Structures (A+), TOEIC 880, plus open-source contribution to linmo.