Aiden Mak logo

Zokforce

Reading lens
Info

I test clinical voice AI before real users touch it — simulated clinic audio, bilingual test plans, honest defect reports.

Role
Internship
AI Engineering Intern
Volunteer Jan 2026 · Intern May 2026 – Present
Stack
Python
STT/ASR
TTS
QA
Proof
clinsim: de-identified case briefs → labeled multi-speaker clinic audio in 5 languages (incl. HK code-switching)
stt_uat: cross-provider TTS generation + accuracy scoring under configurable acoustics
Bilingual (EN + 繁中) UAT plans and a maintained defect log for two AI products
Clinical speech-to-text fails in ways manual testing can't cover — accents, code-switching, noisy rooms.
01 — What Can Be Shown02 Films
Film — clinsim.mp4
01A Clinic in a Simulator

clinsim converts de-identified clinical notes or case briefs into labeled multi-speaker audio — a repeatable clinical encounter for STT pipeline testing, with no PHI anywhere in the loop.

The clinsim simulator runs a scenario
Film — qa_workflow.mp4
02Triage, Reproduce, Verify

User-facing issues get triaged, reproduced and verified with senior engineers — structured defect tracking that turns individual bug reports into release confidence, across English and Cantonese test flows.

The QA workflow, step by step
02 — Business ValueWhat Changes
Privacy

Test coverage without PHI: simulated encounter audio is synthesized from de-identified sources, so realism never costs confidentiality.

Coverage

English and Cantonese encounter flows, plus the long tail of failure modes manual UAT alone can't reach.

Reliability

Reproducible failure cases and documented verification — what release confidence is actually made of.

Internship work — no client systems, screens or data appear on this page; only tooling and workflows within Aiden's own scope.