An auto-grading flashcard agent for CompTIA practice

The scenario

Studying flashcards by yourself has one weakness: you grade your own answers, which means 'close enough' becomes 'correct.' A custom bot that takes your spoken or typed answer to a flashcard, compares it to the canonical answer, and grades you honestly closes that loop. This is the second-most-used Workshop 3 build for participants taking CompTIA exams.

The prompt

You are an auto-grader for CompTIA flashcard practice. I will give you a flashcard (Q + canonical A) and my answer. Grade my answer.

Output format (exactly):

GRADE: correct / partially correct / incorrect
WHY: [one sentence. If "partially correct," name what was missing. If "incorrect," name the most likely confusion.]
EXAM-LEVEL FEEDBACK: [one sentence. What would a real CompTIA exam item-writer say about my answer? Would it pass on the exam?]

Rules:
- Be strict on technical specifics. "DHCP gives out IP addresses" passes; "DHCP is a network thing" fails.
- Be lenient on phrasing. The exam is multiple-choice, so we are practicing recognition, not the wording.
- When my answer reveals a confusion (e.g., I confused HTTP with HTTPS), call out the specific confusion. That is the most useful coaching the agent can give.
- Do not invent acronyms or definitions. If you are unsure whether my answer is correct, say so and ask me to check the source. Do not bluff.
- Do not be condescending. I am paying for this bot to make me better at the exam, not to feel good about wrong answers.

Flashcard:
Q: [paste]
A (canonical): [paste]

My answer:
[paste]

The output

An example grading run.

Q: A user reports that they can ping their default gateway by IP address but cannot reach external websites by name. What protocol should you check first?
A (canonical): DNS.

My answer: I'd check DNS, since the user can ping by IP but not by name, which means name resolution is broken.

GRADE: correct
WHY: Identified DNS as the protocol responsible for name-to-IP resolution and connected the symptom (works by IP, fails by name) to the right diagnosis in one breath.
EXAM-LEVEL FEEDBACK: Strong answer. On a multiple-choice item, you'd land on the right option. The reasoning makes the answer defensible if you got the question half-remembered.

Q: A user reports that they can ping their default gateway by IP address but cannot reach external websites by name. What protocol should you check first?
A (canonical): DNS.

My answer: HTTP, because they can't reach websites.

GRADE: incorrect
WHY: HTTP is the protocol used once a website's address has been resolved; the failure here is that the address can't be resolved at all (works by IP, fails by name). The protocol that resolves names to addresses is DNS.
EXAM-LEVEL FEEDBACK: This is a classic Network+ confusion. A multiple-choice item with "DNS" and "HTTP" as two options would catch this answer. Worth re-reading the OSI model layer 7 mapping for HTTP versus the supporting role of DNS.

Why this earns its keep.

The “wrong answer with the right confusion called out” output is what turns flashcard practice into actual learning. Without the grader, a learner reads the canonical answer, thinks “yeah I knew that,” and moves on. With the grader, the learner sees “you confused HTTP with DNS” — a specific, named confusion. That is the kind of feedback that sticks.

Worth using daily for the four weeks before the exam.

One reasonable answer. Your run may differ. Read it against the scenario before using any of it.

What to watch for

AI graders sometimes mark technically wrong answers as 'partially correct' because the answer 'sounds right.' Push the prompt to be strict. Re-prompt if the agent gets soft.
If you do not paste the canonical answer, the agent grades against its own knowledge — which can be wrong about specifics. Always paste the canonical answer.
Do not paste flashcard content from a paid product (Mike Meyers, Sybex, Jason Dion, CertMaster). Most are licensed; reposting violates the license. Use the AI-generated flashcards from the Workshop 2 activity.
The agent's exam-level feedback is its weakest output. CompTIA does not publish how exam items are scored, so this is the agent's best guess. Treat it as a hint, not a verdict.
Audio-in or speech-to-text grading sounds great but adds privacy and accuracy issues. Type for the workshop; consider voice later, with a non-shared bot.

← Back to agentic use cases