Skill map01
Each row points to the section below that shows it.
| Skill | Where | Proof |
|---|---|---|
| Full-stack and desktop apps | JobbyMail | a Tauri and React front end over a Python pipeline, packaged as a Windows installer |
| Backend services behind payments | NumIsToken | redemption History and Detail services on the backend of a live token flow |
| Reliability and recovery | NumIsToken, JobbyMail | redemptions resumable from the last confirmed state; an append-only event log that makes every rerun safe |
| Performance profiling | NumIsToken | redemption actions cut from ~2–3 s to ~300 ms by removing repeated lookups |
| Debugging intermittent failures | Kigumi Group | two AI-assistant failure classes separated and each pinned with a test |
| Testing and coverage | Kigumi Group, JobbyMail | coverage raised from 90% to 98%; 417 tests across Python, UI, the Rust shell and the browser |
| Handling untrusted input | JobbyMail | HTML email rendered in a sandboxed frame with an allow-list sanitiser and a content policy |
| Entity resolution at scale | JobbyMail | q-gram blocking skips 99.99% of 17.5M candidate pairs before matching |
JobbyMail · a local-first desktop app02
A Windows desktop app I built end to end, Jul 2026 – present: a Tauri and React front end over a Python pipeline that reads a mailbox read-only and keeps a job-application table up to date by itself.
- 01
Built an incremental, read-only mail sync (IMAP with verified TLS, the app password kept only in the Windows credential store) that fetches only new messages, de-duplicates by Message-ID, and stores one odd message in raw form instead of stopping.
- 02
Resolved each email to an application with entity resolution: names cleaned, candidates found by q-gram blocking, then matched on q-gram overlap across the company name, the sender's name and address. Blocking skipped 99.99% of 17.5 million candidate pairs.
- 03
Derived every status from an append-only event log, so running the pipeline twice changes nothing and a wrong guess can be corrected without losing history.
- 04
Rendered untrusted HTML email safely: an allow-list sanitiser, a sandboxed frame that cannot run scripts, and a content policy inside the frame, with remote images blocked until the reader allows them.
- 05
Wrote the tests during the build and ran them all at the end: 417 across four suites, all passing.
NumIsToken · backend behind a payments flow03
Full-Stack Software Engineering Intern at NumIsToken, Sep 2025 – Jun 2026, on the backend behind token redemptions.
- 01
Profiled the redemption lifecycle APIs before changing anything. The two to three seconds per action came from the same lookups and checks, repeated across a single request.
- 02
Removed the redundant calls and measured again: about 300 ms per action, roughly 85% faster.
- 03
Persisted every step of a redemption behind History and Detail services, so an interrupted one can be found and resumed from its last confirmed state without acting twice on the same payment.
Kigumi Group · root-causing recurring failures04
Backend Software Engineer Intern in Hong Kong, Jun – Sep 2025, on KiguLab, a digital wellbeing and AI-ethics curriculum for ages 8–21.
- 01
Started from AI assistant failures that looked random from outside: the same question would work once and fail the next time.
- 02
Separated them by targeted debugging into two classes with different fixes: unstable backend connections, and malformed queries sent to the GPT and image models.
- 03
Measured coverage before and after (90% to 98%, so the untested code shrank by 80%) and pinned each diagnosed failure with a test.