Lab 17 · AI Chatbot Project Lab: Study-Assistant Chatbot Design

🎯 Objective

Build the professional design package for a chatbot — the part real AI projects spend most time on — and evaluate it like a product team.

🧩 Skills You Will Practice

• System prompt and persona design
• Prompt templates and guardrails
• Test-case design
• Failure analysis and iteration

🧰 What You Need

• One AI chatbot (free) to prototype with
• A design document (notes or doc)
• One subject you will tutor (for example Python basics)
• Your Lab 5 prompt templates

📋 Step-by-Step Tasks

1. Define the product in one page: who it helps, the single job (explain + quiz one subject), what it must never do (write graded assignments, invent references), and its tone. This page controls every later decision.
2. Write the system-style instruction you will paste at the start of a session: role, subject scope, teaching method (explain, example, then quiz), format rules, and the honesty rule (say when unsure).
3. Build 4 reusable prompt templates: Explain a topic, Work an example, Quiz me (questions first, answers hidden), Review my answer. Each template has blanks for topic and level.
4. Create a test set of 15 cases before testing: 5 normal study questions, 4 tricky or ambiguous, 3 where the correct behaviour is to refuse or redirect (do my assignment, give exam answers for a live test, personal data request), 3 factual questions you can verify.
5. Run all 15 cases, log for each: Pass / Partial / Fail, what actually happened, and the exact instruction change that would fix a failure. Change the instruction, re-run the failed cases only.
6. Score the final design: pass rate out of 15, the worst remaining failure, and the limitation you would warn a real user about on day one.

✅ Expected Result / Deliverable

A chatbot design package: 1-page product definition, session instruction, 4 templates, 15 logged test cases with fixes, and a final pass-rate with honest limitations — a real AI portfolio artefact.

⚠️ Common Mistakes

• Testing only easy questions the chatbot will obviously pass
• No refusal cases in the test set
• Declaring the chatbot ready without a failure log