Knowledge Bases

Agents are only as good as the knowledge they can access. Without a governed KB, teams duplicate content, index stale wikis, and agents mix conflicting answers from random Slack threads.

Wikipedia with an editorial board, access badges, and a 'last verified' sticker on every article — not a pile of sticky notes in everyone's desk drawer.

Scenario

Engineering publishes a new incident runbook. The KB webhook triggers re-chunking and re-embedding; within 10 minutes the on-call agent retrieves the updated steps instead of last quarter's playbook.

Solution

In Enterprise AI, apply Knowledge Bases to this scenario: Engineering publishes a new incident runbook. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Do this before moving to the next module — reading alone is not enough.

Open the Code Walkthrough below and run it locally. Change one parameter related to Knowledge Bases (e.g. model, temperature, top_k, or tool name), observe the difference in output, and write 2–3 sentences explaining what changed.

Highlighted lines show where Knowledge Bases happens in the code.

1// KB publish webhook handler2app.post("/kb/publish", async (req) => {3  const doc = await kb.upsert({4    id: req.body.id,5    title: req.body.title,6    body: req.body.markdown,7    owner: req.body.owner,8    classification: req.body.classification,9    version: req.body.version,10    published_at: new Date(),11  });12  await ingestionQueue.enqueue({ docId: doc.id, action: "reindex" });13  await audit.log({ event: "kb.publish", docId: doc.id, actor: req.user.id });14});

• Treating Knowledge Bases as a black box without evaluation
• Ignoring cost and latency in production
• Skipping error handling for knowledge bases

Quick recap — the most important points from this module.