Linus Shyu

中文

Linus Shyu (许发鑫)

Student Researcher · Applied ML · Builder · Tech Creator

Email: 0x11@linusshyu.dev

Location: Qinghai, China

GitHub: Linus-Shyu

Bilibili: Linus_Shyu

About Me

I am Linus Shyu (许发鑫), a high school student from Qinghai, China, passionate about computer science, machine learning, and product building. I serve as Co-Founder & CTO at FeelFlow.

I build and ship real-world software while exploring applied machine learning, agent systems, and competitive programming. I share what I learn on Bilibili and YouTube, and document projects and contest writeups on my blog.

My work spans hackathons, open-source tools, and shipped products — I care about ideas that are both technically solid and useful in practice.

News

Research Interests

Applied ML, Agent Systems, Efficient Learning

I study how machine learning systems can stay practical — fast enough to deploy, reliable enough to trust, and structured enough to improve over time. Topics include emotion-aware applications, tool-using agents, and data-efficient modeling.

Competitive programming and hackathon-style engineering keep my implementation skills sharp and grounded in real constraints.

Publications & Preprints

Under Review ATEP at Small Scale: Cache-First Process-Trace Evaluation for Multimodal Tutoring Agents

Submitted to COLM 2026 Workshop on Methods and Opportunities at Small Scale (MOSS).

Cache-first process-trace evaluation for tutoring agents, with deterministic falsifiers, zero training FLOPs, 720 verifiable episodes, and 5,778 checkable frames.

Under Review ATEP: Process-Level Measurement for Steerable Multimodal Tutoring Agents

Submitted to COLM 2026 Workshop AIMS: AI Measurement Science: Toward Rigorous AI Evaluation.

Process-level measurement protocol for multimodal tutoring agents, separating answer correctness, trace grounding, steering behavior, and judge preference.

Experience

Star Lake Explorer AI · Chief Technology Officer

Ongoing — Leading technical development and strategy

FeelFlow · Co-Founder & CTO

Ongoing — emotion-aware product & full-stack engineering

StarFetch · Creator & Maintainer

Ongoing — Rust-based system information tool with Homebrew integration

Apple Swift Student Challenge 2026 · Competitor

2026 — Submitted App Playground project Composer

AdventureX 2026 · Hackathon

2026 — Finalist

Nankathon · Nanjing University

2026 — Participant in hackathon event

MoonShot48 · Beijing

2026 — 3rd Prize & Finalist; team-based product hackathon

42 Wolfsburg · Intensive Coding Program

2026 — Admitted for four-week technical program in Germany

AdventureX 2025 · Hackathon

2025 — Finalist & 3rd Prize (won track with XTerminal, with MSA Beijing)

Google I/O Connect China 2025 · Developer Event

2025 — Guest

Apple Developer Center · Cupertino

2025 — Attendee at in-person developer event

Changsha Hackathon · Hackathon

2025 — Participant

NOI Winter Camp · Competitive Programming

2024 & 2025 — Finalist

Competition Highlights

Education

High School, Qinghai, China (ongoing)

Focus: computer science, mathematics, independent research & projects

Open Source

StarFetch_Core · ★ 12

Linus-Shyu/StarFetch_Core · Homebrew Tap

Rust NeoFetch-style system info tool with adaptive ASCII art and clickable terminal links; macOS / Linux / Windows.

dot-crypto-ticker · ★ 4

Linus-Shyu/dot-crypto-ticker

Lightweight Rust app showing real-time crypto prices on Dot devices.

Luogu-problems · ★ 2

Linus-Shyu/Luogu-problems

Luogu OJ solutions with AC code and writeups.

facial-recognition · ★ 2

Linus-Shyu/facial-recognition

Python facial recognition experiment.

StarDao · ★ 2

Linus-Shyu/StarDao

Solidity smart-contract learning project inspired by RandomDAO.

More on GitHub @Linus-Shyu