Guided browser workshop / Intermediate / Estimated 60–90 minutes
Build a Knowledge Assistant
Retrieve supplied evidence and distinguish grounded answers from unsupported questions.
What you’ll build
A bounded sample workflow to practice retrieval, grounding, source verification. These browser workshops are complete guided exercises, not production implementations.
Prerequisites: basic familiarity with inputs, outputs, and validation. No installation, account, or API key is needed for the sample demonstration.
- Sample input
- Permitted tool or rules
- Validation
- Reviewed output
Evidence inspector / interactive simulation
Supported examples: support reply, message approval, and record retention. Answers use supplied passages only; this is not semantic search or live AI.
Step-by-step workshop
- Read the three supplied policy passages in the demo. No live search is performed.
- Choose a supported question and select the passage that actually supports the answer.
- Return the answer with the document identifier. Abstain if the selected evidence does not support it.
- Test an unrelated question and a misleading passage. Confirm neither produces an invented answer.
Challenge task
Require evidence from the correct policy before answering.
Expected output
A verified sample result for supported inputs; a clear review or failure state for missing evidence, invalid data, or configuration errors. Nothing sent, saved, or changed externally.
Extension idea
Build a local Python lookup over your own fictional passages, then evaluate retrieval separately from answer quality.
Related lesson: Tools, Instructions, and Context