Build an AI evaluation notebook

Let’s build.

A paper or spreadsheet activity. No personal data, paid AI account or programming is required; an educator can supply example answers.

1. 01

   Choose five STEM questions with answers you can check in a textbook or official reference.

2. 02

   Write the expected facts and your reference before examining any generated answer.

3. 03

   Collect one answer per question using the same instructions, or use educator-supplied samples. Record the tool and date if applicable.

4. 04

   Mark each answer for factual correctness, clarity and unsupported claims. Keep these criteria separate.

5. 05

   Explain which errors matter and how a learner should verify the answer. Do not generalise five examples into a claim about overall model accuracy.

# iSpark AI Evaluation: Automated Accuracy Scorer
benchmarks = [
    {"question": "Ohm's Law: 5V across 1000 ohms current?", "expected": "0.005 A (5 mA)"},
    {"question": "ESP32 default logic operating voltage?", "expected": "3.3V"}
]

def evaluate_response(response, expected):
    return expected.lower() in response.lower()

for item in benchmarks:
    print(f"Testing benchmark: {item['question']}")