iPAS Certification Practice Site (AI Application Planner & AIoT Application Engineer)

Overview

An online practice site for the iPAS certifications run by Taiwan’s Ministry of Economic Affairs, now covering two certifications, seven subjects and 2,222 questions: AI Application Planner at Associate and Intermediate level, plus AIoT Application Engineer at Associate level (IoT track).

Core experience: built on the official past exam papers, it offers three modes — Mock Exam, Drill Practice and New Bank Practice — and uses per-option analyses, a mistake notebook and per-topic statistics to create a repeatable learning loop.

Navigation is layered as certification → level → subject → mode, keeping the two certifications’ banks completely separate rather than mixed together.

Subjects Covered

The two AIoT subjects have no mock exam yet: the official prospectus states only “100 points per subject, 70 to pass” without publishing the question count or per-question weighting, so the AI Application Planner’s 50 × 2 rule is not borrowed as a guess.

Question Bank

The latest additions include 100 official questions from the 115-2 AI Application Planner Associate exams and 150 new AIoT topic questions. The source counts above were checked against the question data on September 22, 2026.

Key Features & Experience

• Mock Exam: three repeatable 50-question papers per supported subject, with the same questions when reopening a given paper; timed and scored on submission; 2 points each, 100 total, 70 (35 correct) to pass.
• Drill Practice: work through the original bank in paper order, with instant correct/incorrect feedback and explanations, untimed; filter by all / wrong / unanswered, jump to a question, resume progress or reset it.
• New Bank Practice: the new bank classified by official assessment topic, where every question carries a per-option analysis explaining why the answer holds and where each of the other three options fails.
• Topic Performance: lists “answered” and “correct” per assessment topic, so weak topics surface instead of hiding behind an overall score.
• Mistake Notebook: wrong answers from submitted mock exams are collected automatically for focused review through the drill filter.
• Study Topics: official study-guide summaries for all five AI Application Planner subjects, plus exam-prep notes for the 15 AIoT assessment topics with comparison tables, formula blocks and step flows, plus a 120-entry abbreviation reference, a topic index with full-text search, expand/collapse all, a pre-exam cram mode, per-topic jump-into-drill links and read-progress marks.

Sources & Quality Controls

• Transparent bank provenance: the official PDFs stay in a private, un-versioned directory; pdftotext converts them to markdown and npm run parse:papers produces the committed JSON. Because the output is versioned, re-running the parser and checking git diff --exit-code is a ready-made regression test.
• Hand-written layer kept separate from machine output: explanations and per-option analyses live in their own hand-written layer, so fixes survive a re-parse. Reading the official files question by question also turned up three defects in the official PDFs themselves (a question whose options were replaced by explanation text, an answer contradicting its own reasoning, and an explanation answering a different question) — all fixed and tracked in docs/coverage/bank-defects.md.
• AI-authored questions are identified: the topic-classified questions were written by an LLM and still await human fact-checking; the bank files and project docs say so plainly. The authoring policy sets floors for cognitive level and cross-topic coverage, written into tests as a contract so regressions are blocked.
• CI tests before it builds: GitHub Actions runs the full suite ahead of the build, so the guards on bank integrity, explanation coverage, figure transcription, simplified characters and topic quotas actually gate the deploy.