About

01 / ABOUT

2,271failed rows. One root cause: an empty field

87.5%acceptance pass rate. The rest is on the page too

AllowAny→ default-deny. Wrong order opens everything

argmax(-1)wrong axis. It looked like training. It meant nothing

text-basecollided with a Tailwind scale word. White on white

SEC-001‑030closed one by one before launch

728endpoints, every one actually exercised

cache_keymissing the identity dimension. Cache is authorization

03:47a commit timestamp I should not have been awake for

5,256 → 230lines. Order detail, after the split

no Celeryno Redis, no scheduler. The diagram draws the absence

10 MBof self-hosted fonts. The LCP nail

Every plate on this wall is something that actually happened in one of my repos. Refresh for a different set.

I'm binbin — still a student, and since June 2025 a full-stack developer at Manience Inc., where I build the AI nail platform and the dozens of Django microservices behind it.

During acceptance on the HR system I found that the payroll API cached without the user in the key — inside one TTL window an employee would hit an admin’s cached response and see company-wide salaries. It threw no error, it was not slow, the tests were green. That is when I started believing that cache design is authorization design.

02 / Experience

2025.06 — PRESENT

Full-Stack Developer — Manience Inc.

Full-stack design and development for the AI nail platform and the Django microservices platform — from database schemas and APIs to frontends and AI integration.

— PRESENT

Undergraduate student — NKUST — Dept. of Intelligent Commerce

Finishing my degree while shipping real systems — the alumni platform and the equipment borrowing system both run in production on campus.

2023

Trainee · presented on behalf of CDRI — DIGI⁺Talent Cross-domain Digital Talent Program

Took part in the MOEA DIGI⁺Talent program and, with the EcoBǎo food-surplus marketplace project, presented the results on behalf of the Commerce Development Research Institute (CDRI).

2024

Participating student · full-stack + AI model — National Science and Technology Museum × MOE 2024 AI Health Application Program

Took part in the museum × MOE AI Health Application program with “Four Times for Cook”, a food-therapy recommendation and health-management platform — independently building the Django backend, the React frontend and an entity-tagging recommendation model fine-tuned on Google’s pretrained BERT.