RelayVano

← All postcards

Friday Desk Reset: Turn 200 Notes into a Knowledge Base

Field trips EN

Your deployment checklist is in a chat thread, a useful database fix is in a text file, and last month’s architecture decision is somewhere in your meeting notes. The problem is not a shortage of information. It is that finding an answer requires remembering where you left it.

Use this numbered walkthrough to build a small, searchable personal knowledge base. Start with one active project and a 90-minute session rather than reorganizing your entire digital history.

1. Choose three questions your notes should answer

Open a blank document and write three questions you expect to ask next week. For a developer, these might be: “What is our release procedure?”, “Why did we choose this queue?”, and “Which fixes have worked for connection timeouts?”

These questions define your first collection. Include material that helps answer them; leave unrelated bookmarks and old projects alone. A useful knowledge base starts with a retrieval problem, not a perfect folder structure.

2. Collect source material without deleting originals

Create a local folder named knowledge-base, with four subfolders: inbox, notes, sources, and archive. Export relevant documents and copy loose files into inbox. Save chat excerpts with their surrounding context, not just the sentence that looks important.

  • Keep original exports in sources.
  • Put rewritten, searchable notes in notes.
  • Move processed inbox items to archive.
  • Record the original location and capture date for each item.

For material you cannot export, create a short pointer note containing its title, location, owner, and a summary. Respect workplace access rules; a personal collection should not become an unauthorized copy of restricted information.

3. Turn one source into one reusable note

Choose a source that answers one of your three questions. Rewrite it as a plain-text or Markdown note that makes sense without reopening the original. Use this repeatable layout:

  • Title: a specific problem, procedure, or decision.
  • Summary: two sentences stating the useful answer.
  • Details: steps, constraints, examples, or reasoning.
  • Source: original file path or document location.
  • Status: draft, verified, or outdated.
  • Reviewed: the date you last checked its accuracy.

Name the file for the subject, such as deployment-rollback-procedure.md. Avoid titles like “Useful stuff.” Split a source into multiple notes only when its topics would be retrieved separately.

4. Use AI to draft, not certify, your notes

For repetitive cleanup, ask an AI model: “Convert this excerpt into the note layout below. Preserve commands and dates exactly. Separate explicit facts from inferred explanations. List missing information. Do not invent steps.” Paste the layout and one source excerpt underneath.

If you already work through an OpenAI-compatible SDK, RelayVano is a practical pay-as-you-go option for this task, with an API base of https://api.relayvano.com and models such as Claude Sonnet 5.5. Sign up when you are ready to test a small batch rather than committing your whole archive.

Before sending anything, remove credentials and personal data, confirm you are allowed to share the content, and review the applicable data-handling terms. Compare every generated note with its source. Keep unverified procedures marked draft.

5. Connect notes with explicit relationships

Add a related-notes line when another note provides a prerequisite, exception, or explanation. For example, a deployment checklist can point to the rollback procedure and the decision to use rolling releases.

Create one index document listing your three original questions and the notes that answer them. Add a few consistent tags only if they help filtering, such as deployment, database, and decision. Avoid building a taxonomy before you know what you search for.

6. Test retrieval with yesterday’s problem

Close your source applications. Using only your knowledge-base folder and ordinary text search, answer a recent work question. Try the words you would naturally type, including an error fragment or an alternative term.

If the right note does not appear, improve its title or add the missing terminology. If it appears but cannot support an answer, restore the missing context. You do not need a vector database until simpler retrieval demonstrably fails.

7. Schedule a ten-minute weekly repair

Each Friday, process a few inbox items, check one frequently used procedure, and mark superseded notes outdated with a pointer to their replacement. Back up the folder using an approved service or private repository, keeping secrets out of version history.

Measure progress by whether you can recover a reliable answer quickly—not by how many notes you have collected.

More postcards from this route