https://github.com/tonypeng1/my-wiki-public
AI-powered personal wiki: raw documents → concept articles, summaries, Q&A, and slides, maintained by LLMs and viewed in Obsidian
https://github.com/tonypeng1/my-wiki-public
claude knowledge-base llm obsidian pdm personal-wiki rag second-brain
Last synced: 5 days ago
JSON representation
AI-powered personal wiki: raw documents → concept articles, summaries, Q&A, and slides, maintained by LLMs and viewed in Obsidian
- Host: GitHub
- URL: https://github.com/tonypeng1/my-wiki-public
- Owner: tonypeng1
- Created: 2026-05-16T08:05:39.000Z (about 1 month ago)
- Default Branch: main
- Last Pushed: 2026-05-31T03:27:32.000Z (17 days ago)
- Last Synced: 2026-05-31T05:09:02.044Z (17 days ago)
- Topics: claude, knowledge-base, llm, obsidian, pdm, personal-wiki, rag, second-brain
- Language: Shell
- Homepage:
- Size: 49.8 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# my-wiki
A **personal health wiki** driven by AI prompts. Raw medical source documents (lab results, imaging reports, clinical notes, medication records) are ingested and transformed into structured concept articles, summaries, Q&A records, and slide decks — all cross-linked using Obsidian-style `[[backlinks]]`.
> **Scope: medical data only.** The project's tagging taxonomy, article formats, and MOC structure are purpose-built for clinical domains. Non-medical documents (finance, recipes, general notes) cannot be meaningfully tagged and should not be ingested here. Use a separate wiki for other domains.
## Inspiration
The structure and ideas behind this project are inspired by [Andrej Karpathy's post on LLM Knowledge Bases](https://x.com/karpathy/status/2039805659525644595) (April 2026), in which he describes using LLMs to incrementally compile a personal wiki from raw source documents — with summaries, concept articles, backlinks, Q&A, Marp slides, and health-check linting — all maintained by the LLM and viewed in Obsidian.
## Directory Structure
```
raw/ → source documents (never modify these)
wiki/
home.md → vault entry point; links to all MOCs
index.md → master index of all wiki content
processed.log → list of already-processed raw/ files
summaries/ → one .md per source document
concepts/ → one .md per concept/topic
mocs/ → one moc-{domain}.md per clinical domain
queries/ → saved Q&A outputs
_handoff/ → clean versions intended to be given to someone
_superseded/ → answers replaced by a newer query
slides/ → Marp slide decks
maintenance/ → health check and synthesis reports
sessions/ → transient session scratch pad (not wiki content)
current.md → active session full Q&A log (deleted on close)
log.md → compact session summary, one entry per turn (deleted on close)
archive/ → closed sessions saved as YYYY-MM-DD.md + YYYY-MM-DD-log.md
scripts/ → automation scripts (Python pre-filters for lint prompts, sync helper)
memory/ → persistent facts and corrections used by Claude
.claude/
commands/ → slash command definitions that power the workflows
prompts/ → reusable AI prompt files
CLAUDE.md → project conventions auto-loaded by Claude Code each session
```
## Workflows
This project runs in **Claude Code**. Workflows are invoked as slash commands — type `/` followed by the command name in the Claude Code chat, e.g. `/ingest-first`.
| Prompt | Slash Command | Purpose |
|--------|---------------|---------|
| `p1-first-ingest.md` | `/ingest-first` | Compile the wiki from scratch using all files in `raw/` |
| `p2-incremental-ingest.md` | `/ingest-increm` | Add only new files in `raw/` that haven't been processed yet |
| `p3-qa.md` | `/qa` | Answer a one-off question and save the result directly to `wiki/queries/` |
| `p3a-session-qa.md` | `/session-qa` | Ask a question inside a live session — auto-starts the session if none is active; full prior conversation is in context for each follow-up |
| `p3b-session-close.md` | `/session-close` | End the session — saves worthy Q&A turns to `wiki/queries/`, archives the session log, and cleans up |
| `p3c-session-reopen.md` | `/session-reopen` | Restore a closed session from archive back to `current.md` to continue it |
| `p4a-post-ingest.md` | `/post-ingest` | Post-ingest housekeeping — canonicalize tags, check backlinks, refresh MOCs; run after every `/ingest-increm` |
| `p4b-contradiction-check.md` | `/contradiction-check` | Scan wiki articles for contradictions |
| `p4c-deep-check.md` | `/deep-check` | Monthly content quality check — thin/missing articles, missing aliases, new article candidates |
| `p4d-triage-queries.md` | `/triage-queries` | Interactive triage of misplaced files in `wiki/queries/` root |
| `p4-lint.md` | `/lint` | Full quarterly health check — all tasks above plus a written report to `wiki/maintenance/` |
| `p5-slides.md` | `/slides` | Generate a Marp slide deck on a topic from wiki content |
| `p6-weekly-synthesis.md` | `/synthesis` | Summarize what was added to the wiki this week |
| `sync-to-public.md` | `/sync-to-public` | **Maintainer only.** Copy public files (prompts, commands, templates) to the companion public repo and suggest a commit message. Not needed if you are a user who cloned this repo. |
## Getting Started
1. Open a terminal in this directory and start Claude Code:
```
claude
```
2. Drop source documents into `raw/`.
3. Run `/ingest-first` to build the wiki from scratch, or `/ingest-increm` to add only new files.
4. After each `/ingest-increm`, run `/post-ingest` to canonicalize tags, check backlinks, and refresh MOCs.
5. Use `/qa` to ask a one-off question. The result is saved immediately to `wiki/queries/`.
6. For a conversational session with follow-up questions:
- Run `/session-qa your question` — the session starts automatically on the first question.
- Keep asking follow-ups with `/session-qa your next question`; each turn appends a compact summary to `wiki/sessions/log.md` (read for context on the next turn) and the full answer to `wiki/sessions/current.md` (used only when the session closes).
- When done, run `/session-close` — it saves substantive Q&A turns to `wiki/queries/`, archives both session files to `wiki/sessions/archive/` (as `YYYY-MM-DD.md` + `YYYY-MM-DD-log.md`), and removes both `current.md` and `log.md`.
7. Run `/deep-check` monthly for content quality (thin articles, missing aliases, coverage gaps).
8. Run `/triage-queries` as needed to move misplaced files out of `wiki/queries/` root.
9. Run `/lint` quarterly for a full health check with a written maintenance report.
10. Use `/slides` to generate a Marp presentation on any topic covered in the wiki.
## Conventions
- All cross-references use Obsidian-style backlinks: `[[article-name]]` (filename without `.md`).
- Article formats (concept, summary, query, MOC) are defined in `CLAUDE.md` (auto-loaded by Claude Code; also human-readable).
- Never edit files in `raw/` — they are the source of truth.
- `wiki/index.md` is the entry point for browsing all content.
## Tagging System
All concept and summary files use a **closed set of 24 canonical tags**. New tags require explicit justification (no existing tag fits, 2+ articles would use it).
**Clinical domain tags** — each has a corresponding MOC file in `wiki/mocs/`:
| Group | Tags |
|---|---|
| Labs & biomarkers | `biomarker`, `hematology`, `immunology` |
| Metabolic / endocrine | `metabolic`, `glycemic`, `lipid` |
| Cardiovascular | `cardiology` |
| Organ systems | `hepatic`, `genitourinary`, `gastrointestinal`, `respiratory` |
| Musculoskeletal & neuro | `musculoskeletal`, `neurology` |
| Integumentary & sleep | `dermatology`, `sleep-medicine` |
| Sexual health | `sexual-health` |
**Cross-cutting tags** (no dedicated MOC):
`screening` · `imaging-finding` · `clinical-finding` · `medication` · `procedure`
**Imaging modality tags** (summaries only, used alongside `imaging-finding`):
`ultrasound` · `mri` · `ct`
Non-canonical synonyms (e.g. `cardiovascular`, `lab-test`, `renal`) are mapped to their canonical equivalents in `CLAUDE.md` and must never appear in frontmatter.
### What happens if a non-medical document is ingested?
Non-medical documents (financial records, recipes, book notes, etc.) have no applicable canonical tags, so Claude would either silently misfit them into medical tags or invent new ones — both corrupting the taxonomy. The article would also lack the clinical context that makes concept articles and MOCs useful. The correct action is to reject non-medical files at ingest time and direct them to a separate wiki.
## Git Setup
To put this folder under version control:
1. Open a terminal in this folder and initialize a git repository:
```
git init
```
2. Create a `.gitignore` to exclude files you don't want tracked (optional but recommended):
```
echo "wiki/sessions/current.md" > .gitignore
echo "wiki/sessions/log.md" >> .gitignore
```
3. Stage all files and make the initial commit:
```
git add .
git commit -m "Initial commit"
```
4. To back it up to a remote (e.g. GitHub), create a new repository on GitHub, then:
```
git remote add origin https://github.com//.git
git branch -M main
git push -u origin main
```
5. After that, commit changes as usual:
```
git add .
git commit -m "your message"
git push
```
## License
MIT License