NTHU CS · M.S. student

Jui-Chi Yu.

游睿騏

Portrait of Jui-Chi Yu
National Tsing Hua University

Software, IT & AI Engineering

I build software tools, data systems, and AI applications. My recent work spans enterprise IT automation and reproducible AI research.

Work Experience

Micron Technology

Jul–Aug 2026

IT Engineer Intern

Built an AI-assisted web interface for operational log investigation, with event timelines, ranked evidence and query audit trails.

What I worked on

Connected Copilot CLI through MCP tools to Splunk and read-only Oracle data. Added validated query templates and bounded retrieval so the supporting evidence could be inspected.

Airoha Technology

Jul–Sep 2025

Tool Software Engineer Intern

Completed 114 Ethernet switch test cases covering packet transmission, reception, counters, port mirroring and multicast behavior.

What I worked on

Contributed automation scripts, investigated library/API issues with a mentor, integrated changes with Git and reported suspected defects to R&D.

Selected Projects

Research in progress

Oct 2026–present

Evidence-Linked Memory and Document Retrieval

Retrieval, made inspectable
BM25E5PPR
Shared benchmarkSources + latency

Compare keyword, dense and graph retrieval with the same questions.

Built a Python retrieval benchmark across 1,500 public-data questions, comparing keyword, dense, hybrid and graph-based methods with source attribution, latency profiling and failure analysis; implemented versioned retrieval contracts and 20 functional checks.

1,500 questions · reproducible evaluation

Project notes & images

Python, BM25, E5, RRF, PPR, PyTorch

Evaluated held-out HotpotQA and LoCoMo questions with source-level error analysis. Retained the stronger baselines after graph candidates underperformed, and documented the results for subsequent experiments.

Team project

Smart Medical Patrol Vehicle System

One system, connected data
Docker
PostGISSpatial data
Neo4jGraph relations

Spatial information and graph relationships, in a containerized environment.

Designed and implemented a hybrid spatial/graph database architecture using PostGIS and Neo4j, containerized with Docker. The project was selected by a Taichung clinic for a paid collaboration.

Paid clinic collaboration

Project notes & images

PostGIS, Neo4j, Docker

The database architecture combines spatial information in PostGIS with relationships in Neo4j, packaged in a Docker environment.

Academic project

Oct 2026

Virtualization and Live Migration Lab

Systems in motion
VM 01Source host
QMP / NFSRunning workload
VM 02Target host

A reproducible lab for moving and measuring a running Linux workload.

Automated a nested-KVM live-migration lab with QEMU, NFS and QMP; migrated a running Linux workload with 14 ms QEMU-reported VM downtime and built reproducible CPU, network and VM/microVM benchmarks.

Running-workload migration · reproducible lab

Project notes & images

QEMU, KVM, NFS, QMP, Linux, Python, Bash

Measured 14 ms of QEMU-reported VM downtime in a controlled nested-KVM run. Benchmarks used perf, sysbench, iperf3 and a matched CNN workload.

Retrieval systems, database architecture and infrastructure automation.

Let’s talk.

For software, IT and AI engineering opportunities.