One person's engineering lab: I write code, take systems apart, and write down what broke. Everything here is my own work.
Nineteen case studies — each page has its own visual language: colour, type, grid and layout never repeat.
Small systems I built for fun and for testing ideas. All of them run on this machine with sample data — click in and use them.
Eight couriers, eight field sets, eight vocabularies for status. Queries are dispatched per carrier, then mapped onto one state machine (picked up / in transit / out for delivery / delivered / exception), with exception nodes stored separately. One REST surface and one webhook out, fired only on state change. The hard part is not the request — it is field alignment and idempotent retries.
Open the demo →(sample data)
One orders table, one payments ledger, and a human going cross-eyed. Orders are indexed by number, then every order is classified into unpaid / amount mismatch / overpaid / duplicate payment, with four totals: ordered, received, outstanding, discrepancy count. CSV export carries a UTF-8 BOM so Excel opens it cleanly. You can upload your own CSVs and run it.
Open the demo →(upload your own tables)
Search the list page by condition, open each detail page and pull fields (project, type, region, buyer, agency, publish time, deadline, budget), then download the PDF/Word attachment and parse the body for missing fields. Paging and incremental dedupe run on timestamp plus a unique key; failed rows retry on their own. This page shows the field schema and the export format.
Open the demo →(sample data)
Moving a decade-old site to a new machine: read-only backup with per-file hashes, restore in the new environment, take over old URLs with 301s, wire HTTPS with auto-renewal, then walk the page inventory one by one and keep a verification log. Before touching anything: check for encrypted cores or machine-bound licences — and never write to the source.
Open the demo →(process notes)
No framework, no extensions: PHP reads system metrics straight into SQLite, and a single page shows disk, processes, site reachability and response time. Plus a concurrency benchmark — same machine, different worker_connections and opcache settings, compared on average response time.
Open the demo →(benchmark report)
The tools I reach for. Selection is deliberately conservative: fewer dependencies, and if one file solves it, no framework.
Mostly Python, PHP for lightweight sites.
Hand-written HTML / CSS / vanilla JS, no framework.
Linux servers, from install to years of upkeep.
Automating the repetitive parts.
Short write-ups of problems hit while building sites, moving old systems and tuning performance. Implementation and debugging only.
A place for notes and experiments.
Only two things live here: tools I wrote，和post-mortems from taking systems apart — so that next time I hit the same wall, I waste half an hour less.
Every demo runs on local sample data and the pages carry no tracking scripts. Signing up needs an email address only — no phone numbers.
Deliberately plain underneath: static pages plus a little backend, one Nginx, everything self-hosted, no third-party libraries, no CDN.
Updates are irregular and slow — but everything here actually ran.
All 19 cases ship their HTML / CSS / JS with implementation notes. No phone number, unsubscribe anytime.
Sign up free with email →Already have an account? Sign in