My Own AI Knowledge Wiki Built with Claude — A Second Brain That Accumulates Knowledge and Filters Out What’s Wrong (Obsidian)

Can you trust a wiki organized by AI as-is? Build a work wiki yourself with Claude Code and an Obsidian vault, adding source files and line numbers to every sentence and using comparison scripts to automatically catch citation errors, numerical discrepancies, outdated source material, and contradictions across pages. The advanced course covers source materials in PDF, Excel, scans, web clippings, transcripts, emails, and slides; gold-standard question sets and grader scoring; a 300-page scale; and Git-based team workflows (65 lessons in total). Obsidian is used only to open and view the vault; the vault images shown in the course are reconstructed screens based directly on the practice files.

2 learners are taking this course

Level Beginner

Course period Unlimited

Python
Python
Git
Git
Business Productivity
Business Productivity
claude
claude
Python
Python
Git
Git
Business Productivity
Business Productivity
claude
claude

What you will gain after the course

  • Create a rules file that makes Claude Code write wiki pages with the source file, line number, and quoted passage attached to each sentence.

  • A single line of comparison script finds every citation mismatch, number discrepancy, and missing source across the entire wiki, with page and line numbers.

  • Use the original text fingerprint to identify only the pages where the source text has changed, and update only the affected pages when a revised edition comes in.

  • Find inconsistencies where the same item has different values across multiple pages, then determine which is correct based on the original source.

  • When asked about the wiki, it answers with supporting lines. If there is no evidence, it doesn’t make anything up; it says 'no evidence' and provides the relevant source text.

  • Run inspections automatically using collection skills, hooks, refusal rules, and filters, while allowing only humans to change the inspection criteria.

  • View the weekly verification rate, staleness, contradictions, and pending reviews in a single health-check report.

  • Convert PDFs, Excel files, CSVs, scanned images, web clippings, meeting transcripts, emails, and slides into versions that can be cited by line number, and check the original fingerprints, page counts, row counts, and truncated numbers.

  • Hidden instructions in the source text are filtered out and isolated before collection.

  • Separately catch structural issues that contrast checks can’t detect, such as aliases, glossaries, time-specific facts, decision statuses, and deprecated terms.

  • Use a set of questions with known answers to catch regressions in the wiki, and first evaluate AI judges against human judgments to determine what tasks to delegate.

  • Handling a 300-page vault with hierarchical indexes, re-ingesting only changed source text, and keyword search that requires no installation

  • Teams operate by recording each data collection as a single commit, rolling back changes, isolating only changes that need review, resolving conflicts, using an attribute dashboard, and running daily checks.

From a Wiki That Accumulates to One You Can Verify — Trace Every Sentence Back to Its Exact Line in the Source, and Let Machines Catch Errors First — 30 Lessons in the Basics + 35 Advanced Lessons

Can you trust a wiki organized by AI as-is? We’ll build a workplace wiki from scratch using Claude Code and an Obsidian vault, adding the source file and line to every sentence and automatically filtering out incorrect quotations, numbers, outdated source text, and contradictions between pages with a cross-checking script. The advanced course covers original sources in PDF, Excel, scanned documents, web clippings, transcripts, email, and slides; gold-standard question sets and automated grading; scaling to 300 pages; and Git-based team workflows (65 lessons total). Obsidian is used only to open and view the vault; the vault shown in the course is a reconstructed screen that reproduces the practice files exactly.

The screen shows actual terminal output from executed commands and a reconstructed view drawn exactly from the practice vault files—not a recording of the Obsidian app, and this is not a course on how to use Obsidian. This is not a broad survey of Obsidian features or how to build up a wiki. Instead, it shows in actual execution how AI-generated wikis go wrong (numbers change, quotes are fabricated, sources are missing, pages remain unchanged after the original text changes, and pages contradict one another), then builds safeguards to catch each problem, one by one. The checking tools are deterministic scripts that do not call a language model, so the same input always produces the same result. It also does not hide sentences that are wrong in meaning—something the comparison cannot catch—but treats them as something a person must review.

By the end of this course, you will be able to

  • Create a rules file that tells Claude Code to write wiki pages with the source file, line number, and quoted passage attached to every factual statement.

  • With one line of a comparison script, find citation inconsistencies, numerical discrepancies, and missing sources across the entire wiki, by page and line number.

  • Use fingerprints of the source text to identify only the pages where it has changed, and when a revised edition arrives, update only the affected pages again.

  • Identify contradictions where the same item has different values across multiple pages, and resolve them using the original source.

  • When asked about the wiki, it answers with supporting lines. If there is no evidence, it doesn’t make anything up; instead, it says 'no evidence' and identifies the source text needed.

  • Use collection skills, hooks, refusal rules, and filters to automate checks, while allowing only humans to change the criteria.

  • Review the verification rate, staleness, contradictions, and pending reviews each week in a one-page health check report.

  • Convert PDFs, Excel files, CSVs, scanned images, web clippings, meeting transcripts, emails, and slides into transformed versions that can be cited by line number, and check the source fingerprint, page count, row count, and truncated numbers.

  • Filter out and quarantine hidden instructions in the source material before collecting it.

  • Separately catch structural issues that comparison checks can’t detect, such as aliases, glossaries, time-specific facts, decision statuses, and deprecated terminology.

  • Catch wiki regressions with a gold-standard question set, and first evaluate the AI judge against human judgments to determine what tasks to delegate.

  • A 300-page vault handles scale with hierarchical indexes, recrawling only changed source text, and keyword search that requires no installation.

  • Record each collection run as a single commit, and have the team manage it through rollbacks, extracting only changes that need review, resolving conflicts, property dashboards, and daily checks.

Recommended for people like this.

  • Office workers who have started or are planning to start building up work documents in Obsidian (Markdown)

  • Those who have built an LLM wiki but still wonder, “Can I trust what the AI has organized?”

  • Planning, operations, and customer support professionals who use AI to organize team policies, meeting minutes, and manuals, but worry about incorrect information spreading.

Curriculum (13 sections · 65 lessons · approximately 820 minutes)

  • Section 1. When a Wiki Gets It Wrong — Start with the finished result: trace a one-line answer back to its source, five ways an AI-generated wiki can be wrong, setup and folder roles: vault, Python, and Claude Code, who edits which folder

  • Section 2. Writing a Wiki with Sources — Rules file: require a source for every factual statement, Citation format: file, line, and a quoted excerpt, First collection: turn three sets of meeting minutes into wiki pages, Record status in the front matter: verified, unverified, outdated, Send sentences without sources to the review queue.

  • Section 3. Check against the source — let the machine look first — One pass with the comparison script: is the quoted passage in the source? Verify the numbers: do the numbers in the sentence appear in the quoted passage? Review the errors it catches: sentences where deadlines or amounts have changed. Order of correction: have the source reread, then let a person check only the differences. What comparison can’t catch: a sentence with a correct quote but a wrong interpretation.

  • Section 4. When the source text changes — Find outdated pages using source-text fingerprints; when a revised edition arrives: rerun only the affected pages; when two source texts differ: rules for version recency and precedence; find contradictions between pages: the same item, different values.

  • Section 5. Ask the wiki and answer with evidence — Include evidence lines in answers: cite from the wiki back to the original in two steps, make it say “I don’t know” when the answer isn’t in the wiki, feed good answers back into the wiki: only what’s been verified, find gaps in the wiki through question logs

  • Section 6. Safeguards — Skills and Hooks — Collection skill: collect, cross-check, and display status with one command; hooks: automatically run a cross-check when the wiki is edited; only humans change validation criteria: keep rules and scripts intact; weekly health check: verification rate, staleness, and contradictions on one page.

  • Section 7. Move it into your work wiki — Build a vault from your own materials: Copy the framework and gather your first materials, What not to add: Filter out secrets and personal information, Organize a long-running wiki: Record merges and deletions too, Try it yourself: Find the incorrect statement in a new batch of materials

  • Section 8. Maintenance — Maintenance: Monthly Operations Checklist

  • Section 9. Bringing in source material in multiple formats — The converted version is also a source: original and converted text; PDFs: marking page boundaries and checking page counts; PDF tables and truncated numbers; scans and images: OCR text remains pending until verified; five Excel and CSV pitfalls; citing by sheet location; web clippings: source URL and save date; filtering out hidden instructions; meeting transcripts: timestamp footnotes and speaker verification; email threads: removing quoted replies; slides and speaker notes; a one-page conversion verification report

  • Section 10. Strengthening the Wiki Structure — entity pages and alias tables, glossary: what terms mean in this wiki, time-bound facts, status of decision log pages, measuring whether repeated collection causes bloat, finding deprecated terms

  • Section 11. Measure accuracy with numbers — Create a set of questions with known answers, compare regressions after fixes, citation fidelity: recall and precision, score the evaluator first, audit samples with a fixed seed, tally error categories

  • Section 12. Scaling to 300 Pages — Vault and index size at 300 pages, split with a hierarchical index, re-ingest only changed source text, keyword search without installation, measure runtime and tokens

  • Section 13. Team Operations — One ingestion run, one commit, and how to revert; extracting only what needs review from changed sections; what to do when two people edit the same page; building a review dashboard with properties; scheduling daily checks; advanced operations checklist

Practice Environment

The practice ZIP contains the original documents (policies, contracts, meeting minutes, etc.) for the fictional company “Loop Market,” along with five vaults for the basic course (starter, rules, error-seeded draft, completed, and try it yourself) and six vaults for the advanced course (format conversion, source documents, structure, accuracy measurement, 300 pages, and team operations). PDF text extraction and scanned-text recognition use built-in macOS features, and pre-converted files are included so you can follow along on Windows as well. The inspection tools (comparison, stamp, health check, sieve, and hook) use only the Python standard library, and each lesson includes prompts and an answer key. Obsidian is used to open the vault folders, and the vault images shown on screen are reconstructed views of the practice files. Docker and cloud services are not needed.

Prerequisite Knowledge

  • An environment where Claude Code can be installed and run in the vault folder.

  • Being able to copy and run a single command in the terminal (no coding required)

  • Python 3 (for running diagnostic tools)

What this course does not cover

  • How to Install Obsidian and Use Themes and Plugins

  • PKM methodologies (Zettelkasten, MOC, etc.)

  • Building vector databases and RAG systems

  • Automatic integration with daily notes and messengers

Information

The narration for this course was created using a synthetic AI voice. The Claude Code execution outputs and inspection results shown onscreen are from actual runs. Errors in the draft wiki were deliberately planted by a person for educational purposes, as stated onscreen. AI response wording, token counts, and execution times may vary from run to run. All data is fictional and for educational purposes; numbers may vary depending on the execution environment.

Recommended for
these people

Who is this course right for?

  • Professionals who have started—or are planning to—build up their work documents in Obsidian (Markdown)

  • For anyone who’s made an LLM wiki but is wondering, “Can I trust what the AI put together?”

  • Planning, operations, and customer support staff who use AI to organize their team’s policies, meeting minutes, and manuals but worry about incorrect information spreading.

Need to know before starting?

  • An environment where you can install Claude Code and run it in a vault folder.

  • Able to copy and run a single command in the terminal (no coding required)

  • Python 3 (for running the inspection tool)

Hello
This is dorian05202988

Career Verified

A data craftsman with over 20 years of experience in databases and IT. As a DBA and manager of a DB Platform team, I work with database platforms and specialize in MySQL, PostgreSQL, Oracle, MongoDB, SQL tuning, AWS, Terraform, and Python. Drawing on experience in the gaming, finance, and IT services industries, I explain database and cloud architecture. I have published 『Oracle Practical Tuning: Learning SQL Patterns』 and 『Oracle Pro*C Practical Project Guide』, and have presented on serverless architectures, database selection, and multi-database strategies at AWS Summit 2018 and 2019 and DataYa Nolja 2020. Recently, I have been working on slow-query analysis and tuning recommendations using Bedrock LLM, automated DB specification sizing, and query workflows through a DB metadata MCP server and Slack bot. I explain database automation for developers and AI applications for non-developers through hands-on exercises tailored to the audience and objectives of each course. Rather than applying AI suggestions as-is, I teach a decision-making framework that verifies result equivalence, execution plans, and observability conditions, and defers action when the supporting evidence is insufficient. Company operational data and credentials are not included in the course materials.

More

Curriculum

All

65 lectures ∙ (13hr 40min)

Course Materials:

Lecture resources
Published: 
Last updated: 

Reviews

Not enough reviews.
Please write a valuable review that helps everyone!

dorian05202988's other courses

Check out other courses by the instructor!

Similar courses

Explore other courses in the same field!

Limited time deal

$45.10

24%

$59.40