Most people use AI the same way they use Google: ask a question, get an answer, close the tab. The conversation resets every time. The AI never learns who you are.
I wanted something different. I wanted an AI that knows my goals, my history, my patterns, my blind spots — and actually uses that context to give me advice that fits me, not a generic user.
After a few months of iteration, I built a system that does exactly that. It combines Obsidian as a personal knowledge base with Claude as the reasoning layer. Here's how it works and how you can replicate it.
Why Context Is Everything
The limitation of most AI interactions isn't intelligence — it's memory. You spend the first five minutes of every conversation explaining who you are, what you're working on, and what you've already tried. By the time you get to the actual question, you've burned your patience and the AI's context window.
The fix is to stop starting from scratch. If the AI has persistent access to a structured set of notes about your life, goals, and work, every conversation begins from a foundation of context. You ask the question; it already knows the backstory.
Obsidian is the right tool for this because it stores everything as plain Markdown files, which means an AI can read them directly. No export step, no special integration, no API key. Just files.
What the System Looks Like
The vault has a simple folder structure:
10_Periodic Notes/ → Daily and weekly notes
20_Areas/ → Health, finances, relationships, career
30_Projects/ → Active projects with their own notes
40_Career/ → Resume info, target companies, interview prep
50_Learning/ → Study notes, book summaries, courses
60_Reference/ → Guides, gear lists, setup docs, evergreen reference
Every day I write a brief daily note using a consistent template: mood, energy, what I'm working on, habits, what I learned, what I'd do differently. Over time these accumulate into a detailed picture of how I actually operate — not how I imagine I operate.
The weekly note zooms out: themes for the week, wins, struggles, what's shifting.
The Area and Career notes are where the real depth lives. I have notes on my career direction, self-analysis, target companies, behavioral interview prep, and personal goals — all written in my own words, updated as things change.
How Claude Connects to the Vault
This is the part that makes it work. When I open a Claude Cowork session, I give it access to the vault folder on my computer. It can read any file in the vault.
Before asking anything substantive, I run a one-time "onboarding" prompt:
"Read the following files and build a model of who I am: [daily notes from the past week], [self-analysis], [career direction], [goals]. Then summarize what you've learned about me before we start."
Claude reads the files, synthesizes them, and gives me back a surprisingly accurate picture of my current situation. From that point forward, every response in the session is grounded in that context.
The difference in quality is significant. Instead of "here are 10 general tips for job searching," I get "given your background at Amazon, your preference for product-focused engineering, and the companies you've already researched, here's what I'd prioritize next."
What This Has Produced in Practice
I was skeptical of how useful this would actually be before I built it. Here's what it's produced in the past few months:
- A structured self-analysis that articulated my strengths and career misalignments more clearly than years of solo reflection
- A career direction note that reframed my job search from reactive to strategic
- A ranked list of 10 target companies mapped to my specific work style, location constraints, and background
- Behavioral interview prep grounded in real stories from my career history — not generic STAR templates
- A complete hardware and software inventory for my home office with optimization recommendations
- Daily and weekly planning that reflects what I actually care about, not a generic productivity framework
The common thread: the AI isn't generating generic content. It's applying reasoning to my specific situation because it has the context to do so.
The 9-Step Setup
I documented the full setup in a public GitHub repo. The short version:
- Create the vault structure — the folder schema above is a good starting point
- Build your daily note template — mood, energy, non-negotiables, schedule, habits, reflection
- Write organic-first — start by writing freely about your actual situation before structuring anything
- Set up periodic notes — daily and weekly cadence, linked to each other
- Set up Git backup — the Obsidian Git plugin handles this automatically; your notes are version-controlled and optionally cloud-backed
- Get Claude Cowork (or use the API) — you need an AI that can access files, not just a chat interface
- Run your first big conversation — onboard Claude with your key notes, then ask it to help you with something real
- Build domain-specific notes — career, health, projects, whatever matters to you
-
Add source frontmatter — tag your notes with
source: organicvssource: ai-assistedso you always know what came from you
The full repo including daily note template, weekly template, and detailed guide: https://github.com/aschung212/obsidian-ai-coach
The Principle That Makes It Work
The system only works if the vault reflects reality. If your notes are aspirational rather than honest, the AI will give you advice optimized for a version of yourself that doesn't exist.
Write what actually happened. Write what you actually think. Write what you're actually struggling with. The more accurate the input, the more useful the output.
That's the whole thing. The technology is simple. The discipline is the hard part.
If you set this up and run into questions, drop them in the comments. Happy to help.
Top comments (0)