Chapter 1 What This Is
Hey Maya,
I built this guide because I kept getting the same question from people around me: "How do I actually use AI for real work?" Not the TikTok version. Not a 30-second demo. The real thing.
Here's the way I think about it. You know the Barrier Reef back home? From the surface it looks like one big thing. But if you dive in, there are distinct zones. The reef wall is the foundation. The seagrass beds feed the fish. The mangroves protect the shoreline. Each zone does its own job, but they all connect into one living system.
Your AI workflow works the same way. There are distinct pieces (a brain file, a workspace, research tools, planning docs, agent teams) and each one does something specific. But they connect. And once they're connected, the whole system is more powerful than any single piece.
This guide walks through all of it. Not in theory. In practice, with the exact commands and prompts you need.
By the end, you'll be able to:
- Set up an AI assistant that knows your projects, your voice, and your preferences
- Build a workspace that keeps all your side hustle work organized and backed up
- Research any topic deeply using three AI tools that chain together
- Write planning documents that tell the AI exactly what to build
- Run parallel AI agents that work on different parts of a project at the same time
- Maintain all of this with a simple monthly check-in
Time estimate: about two hours to set everything up the first time. After that, it just works.
Hey Mom,
I built this guide for you because I know you're going to crush whatever comes next in your career, and I want you to have the same tools I use every day.
Think of it like starting a hospital shift. You don't walk onto the floor blind. You read the chart. You check the board. You know which patients need what before you touch anything. That preparation is what makes the shift run smoothly.
AI works the same way. If you just open it cold and ask a question, you'll get a generic answer anyone could have gotten. But if you set it up right, with a "chart" that tells it who you are, what you're working on, and how you think, every conversation after that is like working with someone who already did the pre-read.
This guide walks through everything. Interview preparation, research, career planning, even using recording devices to capture meetings and build a personal knowledge base over time.
By the end, you'll be able to:
- Set up an AI assistant that knows your healthcare background, your MBA, and your career goals
- Build an organized workspace for interview prep, certifications, and research
- Research any healthcare topic using three AI tools that work together
- Use voice recordings and transcripts to build a personal knowledge base
- Write planning documents that structure your job search and interview prep
- Run parallel AI agents (one researches the company, one preps questions, one reviews your resume)
Time estimate: about two hours to set everything up. After that, it just works.
Chapter 2 Setting Up
Before you can build anything on the reef, you need to get to Belize. This chapter is the flight, the water taxi, and checking into your spot on the caye. A little boring, but necessary exactly once. Every step below has pictures showing you exactly what to expect.
Before your first shift, you have to get your badge, your access codes, and your locker set up. A little tedious the first day, but you only do it once. Every step below has pictures showing you exactly what to expect.
What you'll install
Four things, in this order. About 20 minutes total.
Node.js is the engine that powers Claude Code. Think of it like this: Claude Code is the car, but it needs an engine to run. Node.js is that engine. Without it, you'll get an error that says "command not found."
Go to nodejs.org in your web browser. You'll see a page that looks like this:
Node.js — Run JavaScript Everywhere
Download Node.js for your platform. The LTS version is recommended for most users.
Download Node.js (LTS) ↓ Current (latest features)Click the green button. That's the stable version. Ignore "Current."
Click the green "Download Node.js (LTS)" button. A file will download to your Downloads folder. It'll be called something like node-v24.14.1.pkg.
Double-click the .pkg file to open the installer. Click "Continue" on each screen, then "Install." It will ask for your Mac password (the one you use to log in to your computer). Enter it and click "Install Software."
When it says "Installation was successful," close the installer. That's it.
Verify it worked
Open Terminal (press Cmd + Space, type "Terminal", press Enter). Then type this and press Enter:
node --version
You should see something like this:
v24.14.1
If you see a version number like v24.x.x, Node.js is installed. If you see "command not found," the install didn't work. Try downloading and running the .pkg installer again.
Common problems you might hit
Problem: "npm: command not found"
This is what happens if you try to install Claude Code before installing Node.js. The screen will look like this:
-bash: npm: command not found
Fix: Go back and install Node.js first (the steps above). npm comes bundled with Node.js. Once Node.js is installed, this command will work.
Problem: A message about "zsh"
When you open Terminal, you might see a message that looks like this:
To update your account to use zsh, please run `chsh -s /bin/zsh`.
Ignore this message completely. It is NOT an error. Your Mac is suggesting you switch to a newer shell, but everything works perfectly fine without switching. Do NOT run the chsh command it suggests. Just continue with the steps below.
Now that Node.js is installed, this will work. Open Terminal and paste this command:
npm install -g @anthropic-ai/claude-code
You'll see a bunch of text scrolling by. That's normal. It's downloading and installing Claude Code. Wait until it finishes (you'll see your cursor come back). It looks something like this:
added 1 package in 12s
$ _
When you see the $ prompt again, it's done. Now launch it:
claude
The first time, it will open a browser window asking you to sign in. If you don't have an account yet, create one at claude.ai (free). After signing in, go back to Terminal. You'll see a text input where you can start typing questions.
For Windows or Linux, check the official Claude Code docs.
VS Code is a free app from Microsoft that makes working with files visual. Instead of typing commands to see your folders, you get a sidebar that looks like Finder but is designed for projects. You can click through folders, open files, and even run Claude Code inside it.
Go to code.visualstudio.com and click the big blue download button.
Visual Studio Code
Free. Built on open source. Runs everywhere.
Download for Mac ↓Click the blue download button. It detects your operating system automatically.
A .zip file will download. Double-click it to extract. Then drag the "Visual Studio Code" app into your Applications folder. Open it from Applications (or search "Visual Studio Code" in Spotlight).
What VS Code looks like when it opens
The first time you open it, you'll see a welcome tab. Close that. What matters is the layout:
Open a folder: Go to File → Open Folder in the menu bar. Navigate to wherever your project lives (for example, your Desktop or Documents folder). Select the folder and click "Open."
Why VS Code matters: visual folder management
Here's the difference. Without VS Code, you manage files by typing commands in Terminal. With VS Code, you get a visual sidebar that shows everything in your project at a glance.
After you set up your workspace (Chapter 4), your VS Code will look something like this:
2
3## About Me
4Maya Rashid. Building a Belize
5travel and lifestyle brand...
6
7## My Project
8Content creation + tourism guides
On the left: every file and folder in your project. Click any file to open it. On the right: the file's contents. This is how you see your whole workspace at a glance.
After you set up your workspace (Chapter 4), your VS Code will look something like this:
2
3## About Me
4Former Director of Radiology,
5RN, MBA. Pursuing senior
6healthcare leadership roles...
7
8## Career Goals
On the left: every file and folder in your project. Click any file to open it. On the right: the file's contents. This is how you see your whole workspace at a glance.
VS Code has an "extensions" system that adds features. The Claude Code extension lets you talk to Claude directly inside VS Code.
- In VS Code, click the square icon on the left sidebar (or press
Cmd+Shift+X) - In the search bar that appears, type "Claude"
- Find "Claude" by Anthropic and click Install
This extension is optional. You can always use Claude Code from the terminal instead. But the extension gives you a nice chat panel right inside VS Code.
Here's the best part: you don't need a separate Terminal app anymore. VS Code has a terminal built right into it. Press Ctrl + ` (the backtick key, usually above Tab) or go to View → Terminal.
2
3Your brain file content
4appears here when you
5click a file on the left.
Welcome to Claude Code! Type your question...
> _
This is your new home base. Three panels in one window:
- Left sidebar: Your folders and files (click to open any file)
- Center: The file you're working on
- Bottom: Terminal where you run Claude Code and other commands
From now on, you'll do almost everything inside VS Code. Type claude in that bottom terminal panel, and you're talking to your AI assistant while seeing all your files.
Other tools (install later)
Gemini CLI (web search AI)
Gemini is Google's AI. The CLI version lets you ask it questions from the terminal, and it can search the live web for answers. This is useful for finding current information that Claude might not have.
Installation instructions are at github.com/google-gemini/gemini-cli. It's free with any Google account.
Don't stress about installing Gemini CLI right now. You can do Chapters 1-7 with just Claude Code. Come back and install Gemini when you reach Chapter 6 (Research).
NotebookLM (research library)
NotebookLM is a free Google tool at notebooklm.google.com. No installation needed. Just sign in with your Google account.
Think of it as a personal research library. You upload documents, articles, YouTube videos, and PDFs. Then you ask it questions and it answers using only your sources, with citations.
Pricing: Start Free
| Tool | Free Tier | Paid Upgrade | When to Upgrade |
|---|---|---|---|
| Node.js | Always free | N/A | Never |
| Claude | Limited daily messages | Pro $20/mo, Max $100/mo | When you hit daily limits and it's costing you time |
| VS Code | Always free, all features | N/A | Never |
| Gemini CLI | Free with Google account | Google AI Premium $19.99/mo | Probably never for most people |
| NotebookLM | 100 notebooks, 50 sources each | Google AI Pro/Ultra = 5x limits | Only if you hit 100 notebooks |
| GitHub | Unlimited private repos | Pro $4/mo | Probably never |
My recommendation: Start 100% free. Every tool here has a free tier that's more than enough to learn and build real things. Upgrade Claude to Pro only when you're hitting the daily message limit consistently and it's slowing you down. Everything else stays free for a long time.
If you're a student, check GitHub Education (free Copilot and extras) and Google education deals before paying for anything.
Chapter 3 Your Brain File
The reef doesn't exist without the foundation. Coral polyps build the base structure, and everything else grows on top of it. Your brain file is that foundation.
In nursing, the chart is everything. Without it, every new nurse on the shift starts from zero. Your brain file is the chart for your AI assistant.
What is it?
A brain file is a single text file called CLAUDE.md that sits in your project folder. Claude reads it automatically every time you start a conversation. It tells the AI who you are, what you're working on, and how you like things done.
Without one, every conversation starts cold. You re-explain yourself every time. With one, the AI already knows your context before you type a word.
Don't write it by hand
The single biggest mistake people make: they try to write the brain file from scratch. Don't. Let the AI interview you and build it from your answers.
Open Claude Code in your project folder and paste this prompt:
I'm setting up this project for the first time. Build me a CLAUDE.md brain file by interviewing me. Ask me these questions ONE AT A TIME. Wait for my answer before moving on: 1. What's your name and what do you do for work? 2. What side project or business are you building? 3. Who is your target audience or customer? 4. What's your main goal with this AI assistant? (content creation, business planning, research, product dev, or a mix) 5. What tools or platforms do you already use? (social media platforms, design tools, writing tools, etc.) 6. What's your voice or style? (casual, professional, educational, funny) 7. What topics are you an expert in? 8. What should the AI NEVER do? (things that annoy you, wrong assumptions) 9. What does "done" look like for a typical task? After all questions, generate a CLAUDE.md file with these sections: ## About Me ## My Project ## Target Audience ## What I Use This AI For ## Tools & Platforms ## My Voice & Style ## Rules (things to always/never do)
Just answer honestly. Talk like you'd talk to a friend. The AI will organize your answers into a structured brain file.
What the result looks like
After the interview, you'll have something like this in your project folder:
# Brain File
## About Me
Maya Rashid. Background in [your field]. Building a side business
while working full-time.
## My Project
A Belize-inspired travel and lifestyle content brand. I create
guides, social posts, and eventually products for people planning
trips to Belize.
## Target Audience
First-time Belize travelers, age 25-40, who want authentic local
experiences instead of resort packages.
## What I Use This AI For
- Drafting social media posts and captions
- Researching travel topics and local businesses
- Building a content calendar
- Writing business plans and marketing strategies
- Brainstorming product ideas
## Tools & Platforms
Instagram, TikTok, Canva, Google Docs
## My Voice & Style
Warm, knowledgeable, personal. I write like I'm texting a friend
who just asked "what should I do in Belize?" — not like a travel
magazine.
## Rules
- Never use generic travel cliches ("hidden gem", "off the beaten path")
- Always write in first person
- Keep captions under 150 words unless I say otherwise
- When I say "draft," give me a complete version, not an outline
- Ask me to clarify before making assumptions about my business
I'm setting up this project for the first time. Build me a CLAUDE.md brain file by interviewing me. Ask me these questions ONE AT A TIME. Wait for my answer before moving on: 1. What's your name and professional background? 2. What are you currently working toward? (new job, career change, certification, research project) 3. What's your area of expertise? 4. What's your main goal with this AI assistant? (interview prep, research, career planning, writing, or a mix) 5. What tools or systems do you already use? (job boards, LinkedIn, specific healthcare platforms, etc.) 6. Do you use any recording devices for meetings or interviews? (Limitless pendant, phone recorder, Otter.ai, etc.) 7. What should the AI NEVER do? (wrong assumptions, things that bother you) 8. What does your ideal prepared state look like before an interview? 9. What does "done" look like for a typical task? After all questions, generate a CLAUDE.md file with these sections: ## About Me ## Career Goals ## Expertise & Background ## What I Use This AI For ## Tools & Platforms ## Interview Prep Approach ## Rules (things to always/never do)
Just answer naturally, like you're talking to someone. The AI organizes your answers into a structured file.
What the result looks like
# Brain File ## About Me [Your name]. Former Director of Radiology, Registered Nurse, MBA. Decades of healthcare leadership experience. ## Career Goals Pursuing a senior healthcare administration role. Open to Director-level positions in radiology, imaging, or operations. ## Expertise & Background - Radiology department management and workflow optimization - Nursing (clinical bedside + administrative) - Healthcare operations, budgeting, compliance - MBA: strategic planning, organizational leadership ## What I Use This AI For - Interview preparation (research companies, practice questions) - Resume and cover letter tailoring per role - Healthcare industry research - Certification tracking and CME planning - Processing meeting recordings into structured notes ## Tools & Platforms LinkedIn, Indeed, healthcare job boards, Limitless pendant for recording practice interviews ## Interview Prep Approach I want to walk into every interview knowing: 1. The organization's recent news and strategic priorities 2. The department's size, structure, and challenges 3. Answers to the 10 most likely questions, tailored to this role 4. Three questions I should ask THEM ## Rules - Never fabricate healthcare statistics or credentials - Always cite sources when presenting research - When I share a recording transcript, extract: key themes, my strong answers, areas to improve, and follow-up actions - Ask me before assuming my experience level on a topic
The Six Golden Rules
Chapter 4 Your Workspace
Ambergris Caye is the main island. Everything connects to it. Your workspace is the same idea: one root folder that holds your entire AI-powered operation.
Example workspace for a side hustle
my-belize-brand/
CLAUDE.md ← root brain file (knows everything)
business-plan/
CLAUDE.md ← knows financials, market, competitors
market-research.md
financial-projections.md
content-calendar/
CLAUDE.md ← knows your brand voice, posting schedule
march-2026.md
april-2026.md
research/
competitor-analysis.md
belize-tourism-data.md
assets/
logos/
photos/
Notice the nested CLAUDE.md files. The root brain file knows the big picture. Each sub-folder brain file knows the specifics of that project area.
When you open Claude Code inside content-calendar/, it reads both the root brain file AND the content-calendar brain file. It knows your overall business AND your specific brand voice. That layering is what makes the outputs good.
Think of your workspace like the floor you manage. The main station has the big picture. Each patient room has its own chart. Your workspace works the same way.
Example workspace for career transition
my-career-hub/
CLAUDE.md ← root brain file (knows your full background)
interview-prep/
CLAUDE.md ← knows your interview approach
company-research/
mercy-health.md
hca-healthcare.md
practice-questions.md
star-stories.md
certifications/
CLAUDE.md ← knows your cert timeline
cme-tracker.md
renewal-dates.md
research/
healthcare-trends.md
salary-data.md
recordings/
transcripts/
2026-03-15-practice-interview.md
2026-03-20-networking-call.md
resume-versions/
director-radiology.md
operations-manager.md
Notice the nested brain files. The root brain file knows your full background. The interview-prep brain file knows your specific approach to interviews. The certifications brain file tracks your renewal timeline.
When Claude is inside interview-prep/, it reads both brain files. It knows your career history AND your interview strategy. That layering is what makes the answers specific to you.
When to nest vs. separate
If two projects share context (like marketing and business plan for the same brand), put them in the same root folder with sub-folders.
If two projects have completely different context (like a cooking blog and a real estate side hustle), give them separate root folders, each with their own brain file.
Let Claude build the structure
You don't have to create all these folders manually. Just tell Claude what you need:
Set up my workspace structure. I need folders for: - [describe area 1] - [describe area 2] - [describe area 3] Create the folders and a CLAUDE.md brain file in each one. Each brain file should know what that folder is for.
Claude will create the folders, write the brain files, and you're ready to go.
Chapter 5 Git — Never Lose Your Work
The water taxi between Belize City and the cayes is the lifeline. If it stops running, you're stuck. Git is your water taxi for files. It moves your work between your computer and the cloud, and it keeps a copy of every version.
Think of this like charting in an EMR. Every entry is timestamped and saved. You can always go back and see what was documented at any point. Git does the same thing for your files.
What git is (plain language)
Git is a save system that keeps every version of every file you've ever saved. Like Google Docs version history, but for your entire folder.
GitHub is the cloud backup. Your work lives on your computer AND on GitHub's servers. If your laptop dies tomorrow, everything is safe.
Public vs. Private
Private = only you can see it. Nobody else knows it exists.
Public = anyone on the internet can find and read it.
For personal projects, side hustles, and anything with your notes in it: always private.
One-time setup
If you don't have a GitHub account, create one at github.com first. Free accounts get unlimited private repos.
# 1. Initialize git in your project git init # 2. Save everything for the first time git add . git commit -m "initial setup" # 3. Create a private repo on GitHub and push # (replace YOUR-USERNAME and YOUR-PROJECT) gh repo create YOUR-PROJECT --private --source=. --push
The gh command is the GitHub CLI. Install it from cli.github.com if you don't have it. After installing, run gh auth login to connect it to your account.
The daily pattern
Every time you finish working on something, save your changes:
# Pull the latest (prevents conflicts) git pull # Save your changes git add . git commit -m "describe what you changed" # Push to GitHub (cloud backup) git push
When to commit
- Finished something? Commit.
- About to try something risky? Commit first, so you can undo if it goes wrong.
- End of the day? Commit and push. Your work is now backed up.
Or just tell Claude: "Commit my changes and push to GitHub." It will do it for you.
Chapter 6 The Research Superpower
The jungle interior of Belize is dense, ancient, and full of things you'd never find on a guided tour. AI research works like having a machete and a local guide at the same time. You can cut through massive amounts of information and find exactly what matters.
Why this matters for your money
Every time you paste a long article or YouTube transcript directly into Claude, you burn through your daily message limit faster (on free) or use more of your paid quota. That's like printing a whole book just to find one paragraph.
NotebookLM solves this. It holds all your research OUTSIDE of Claude. You ask Claude a question, it asks NotebookLM, gets a sourced answer, and barely uses any of your allowance.
The YouTube research pipeline
This is something you can do today, completely free:
- Find videos on your topic (YouTube search, or use Gemini CLI to discover them)
- Add them to NotebookLM at notebooklm.google.com. Paste the YouTube URL as a source. NotebookLM reads the entire transcript automatically.
- Add anything else you find: articles, PDFs, your own notes
- Ask NotebookLM questions. It answers using only your sources, with citations showing exactly where each fact came from.
- Feed those answers to Claude when you need to build something from the research.
Using Gemini CLI for web discovery
Gemini has something Claude doesn't: real-time Google Search. Use it to find things, then add what you find to NotebookLM.
# Find YouTube videos on a topic gemini -p "Find 5 recent YouTube videos about starting a tourism brand in Belize" --yolo # Find current articles gemini -p "What are the latest trends in eco-tourism marketing for 2026?" --yolo
Take the URLs from Gemini's response and add them to your NotebookLM notebook. Now you have a research library that grows every time you find something new.
The continuous research loop
This loop compounds. Each round of research makes the next round smarter because NotebookLM has more context. After a few cycles, you have a research library that would have taken weeks to build manually.
Think of research like doing rounds on a new unit. You read the charts, talk to the nurses, check the labs. AI research is the same process, but you can cover 10x the ground.
Why this matters for your time
When you're preparing for interviews, you need to research companies, departments, recent news, and industry trends. Doing this manually takes hours per company. With this pipeline, you can research a company in 15 minutes and have organized, citable notes.
The research pipeline
- Use Gemini CLI to find current information about a company or healthcare topic
- Add everything to NotebookLM: articles, YouTube talks by hospital leadership, annual reports, job postings
- Ask NotebookLM questions like "What are this hospital's top three strategic priorities?" or "What challenges is this radiology department facing?" It answers with citations.
- Feed those answers to Claude to generate interview prep materials.
# Find recent news about the organization gemini -p "Recent news about [Hospital Name] radiology department 2026" --yolo # Find talks or interviews by their leadership gemini -p "Find YouTube videos of [Hospital Name] CEO or CMO speaking about strategy" --yolo
Add every URL to your NotebookLM notebook for that company. Then ask it: "Based on these sources, what are the three biggest challenges this organization is facing, and how does my background address them?"
Recording + Transcripts: Your Personal Knowledge Base
This is where it gets powerful for interview prep specifically.
What a recording device does
A device like the Limitless pendant is a small wearable that records conversations (with everyone's knowledge and consent). After the conversation, you get a full written transcript. You can also use your phone's voice recorder or an app like Otter.ai for the same purpose.
How to use recordings for interview prep
- Record your practice interviews. Have a friend ask you common questions. Wear the device or use a phone recorder.
- Get the transcript. The recording becomes text that Claude can read.
- Tell Claude to analyze it. It will identify your strong answers, your weak spots, filler words, and areas where you need more specific examples.
- Add transcripts to NotebookLM. Over time, you build a library of your own interview performances. NotebookLM can spot patterns across multiple sessions: "You consistently struggle with questions about budget management" or "Your strongest answers are about team building."
I just had a practice interview. The transcript is in recordings/transcripts/2026-03-25-practice.md Analyze my responses: 1. Which answers were strong and why? 2. Where did I ramble or lose focus? 3. What questions did I struggle with? 4. Suggest better responses for my weak answers, using my actual experience from my brain file. 5. Give me an overall grade (A through F) with specific things to work on before my next practice.
Voice diary / journal use
Beyond interviews, you can use recordings as a daily journal. Talk into your recorder about your day, your thoughts on a job posting, or your career goals. The transcript lands in your workspace. Claude can extract themes, action items, and patterns over time.
These recordings are private. They live in YOUR workspace, backed up to YOUR private GitHub repo. Nobody else sees them unless you share the repo link.
The research loop for healthcare
Each round of research makes the next round smarter. After preparing for three or four interviews, your NotebookLM library becomes a deep resource on the healthcare job market that no one else has.
Chapter 7 PRDs — Tell the AI What You Want First
If you hire a contractor and say "build me a house," you'll get... something. Probably not what you wanted. A PRD (Product Requirements Document) is the blueprint. For AI, it's the same concept: tell it what you want BEFORE it starts building.
A PRD is simple
It's just four questions:
- What's the problem? Why does this need to exist?
- Who is it for? Who benefits?
- What does "done" look like? How do I know it's finished?
- What should it NOT do? Guardrails and boundaries.
The voice-to-text PRD
You don't even need to type this out. Open Claude and literally talk through what you want. Then say "turn what I just said into a PRD."
Plan mode
Before Claude builds anything big, tell it to plan first:
I want to build a content calendar for my Belize travel brand. Before you start building anything, make a plan. Show me: 1. What files you'll create 2. What each file will contain 3. What order you'll do it in Don't start until I approve the plan.
Claude will show you its plan. You approve, redirect, or ask questions. Then it builds exactly what you agreed on. No surprises.
Example: Content calendar PRD
## Content Calendar PRD ### Problem I post inconsistently on Instagram and TikTok. Some weeks I post 5 times, some weeks zero. I need a system that tells me what to post and when. ### Who It's For Me (the creator) and eventually a VA who might help schedule. ### What "Done" Looks Like - A monthly calendar with 3 posts/week - Each post has: topic, platform, caption draft, visual direction - Mix of content types: tips, personal stories, destination guides - Tied to Belize seasons (lobster season, rainy season, holidays) ### What It Should NOT Do - Don't generate actual images (I'll shoot those) - Don't schedule posts (I'll use Canva/Later for that) - Don't include TikTok trends (those are too time-sensitive)
In healthcare, a treatment plan isn't something you wing. You assess, you plan, you execute, you evaluate. SOAP notes follow a structure for a reason. A PRD is the same discipline applied to AI work.
A PRD is simple
Four questions, same as the assessment process:
- What's the problem? (Subjective: What brought you here?)
- Who is it for? (Who benefits from solving this?)
- What does "done" look like? (Objective: measurable outcomes)
- What should it NOT do? (Assessment: guardrails and limitations)
Plan mode
Before Claude builds anything big, tell it to plan first:
I want to build an interview preparation system for healthcare leadership roles. Before you start building anything, make a plan. Show me: 1. What files you'll create 2. What each file will contain 3. What order you'll do it in Don't start until I approve the plan.
Example: Interview prep system PRD
## Interview Prep System PRD ### Problem Each time I prepare for an interview, I start from scratch. I re-research the company, re-write STAR stories, and forget what worked last time. I need a system that builds on itself. ### Who It's For Me. A healthcare professional with 20+ years of experience transitioning into senior leadership roles. ### What "Done" Looks Like - A reusable research template for each target company - A library of 15-20 STAR stories mapped to common questions - A "lessons learned" log from each practice and real interview - Pre-built question banks for: radiology director, ops manager, healthcare admin roles - Recording transcript analysis that tracks improvement over time ### What It Should NOT Do - Don't fabricate experience or embellish my background - Don't generate generic answers (everything should reference MY specific experience) - Don't assume which role I'm preparing for (ask first) - Don't include salary negotiation (I'll handle that separately)
Why this matters
Without a PRD, the AI guesses what you want. With one, it builds exactly what you specified. The 5 minutes you spend writing a PRD saves you 30 minutes of back-and-forth fixing things that weren't right.
You can also just talk it out: open Claude, describe what you want stream-of-consciousness style, and then say "turn that into a PRD." The AI will structure your rambling thoughts into a clean document.
Chapter 8 Agent Teams
Imagine you could hire a small council of brilliant people. One is a market researcher. One is a financial analyst. One is a copywriter. You give them all the same brief, they work in parallel, and a supervisor combines their work into one coherent output.
That's agent teams. Multiple AI instances working on different parts of a problem at the same time.
Think of a code team in the ER. One person handles airway, one handles IV access, one monitors vitals, one documents. Everyone works their role in parallel, and the team lead coordinates. Agent teams work the same way.
The mental model
Single agent = one perspective. It's like getting a second opinion from one doctor. Helpful, but limited by one point of view.
Two agents on the same task = two perspectives. Where they agree, you're probably right. Where they disagree, you need to think harder. This catches blind spots.
Agent team = multiple specialists working in parallel, then a supervisor synthesizes everything. This is for big, multi-part projects.
Level 1: Dual verification (two models, one question)
The simplest version of "agent teams" is just asking the same question to two different AIs and comparing answers:
# Ask Claude your question in Claude Code, then also ask Gemini: gemini -p "Your question here" --yolo # Compare the two answers. Where they disagree = dig deeper.
This takes 30 seconds and catches mistakes that a single AI might miss. Use it for anything important: business decisions, factual claims, strategic advice.
Level 2: Parallel agents in Claude Code
Claude Code can spawn multiple agents that work at the same time. You describe the overall project, and each agent handles one piece:
I need to build a complete business plan for my Belize travel brand. Break this into parallel tasks: Agent 1 (Market Researcher): Research the Belize tourism market. Size, growth trends, competitor analysis, target demographics. Write to business-plan/market-research.md Agent 2 (Financial Analyst): Build financial projections. Startup costs, monthly expenses, revenue scenarios, break-even. Write to business-plan/financial-projections.md Agent 3 (Copywriter): Write the executive summary and pitch. Brand story, value proposition, competitive advantage. Write to business-plan/executive-summary.md After all three finish, combine them into one cohesive business plan at business-plan/FULL_PLAN.md
Three agents work simultaneously. What would take you three sessions in sequence takes one session in parallel.
I have an interview at [Hospital Name] for a Director of Radiology position next week. Break my prep into parallel tasks: Agent 1 (Company Researcher): Research this hospital. Recent news, strategic priorities, department size, leadership. Write to interview-prep/company-research/[hospital-name].md Agent 2 (Question Prepper): Generate the 15 most likely interview questions for this specific role at this specific hospital. For each question, draft a STAR-format answer using my experience from my brain file. Write to interview-prep/questions-[hospital-name].md Agent 3 (Resume Reviewer): Review my resume against this specific job posting. Identify gaps, suggest rewording, flag areas where I'm underselling my experience. Write to resume-versions/[hospital-name]-tailored.md After all three finish, create a one-page interview brief at interview-prep/BRIEF-[hospital-name].md
Three agents working in parallel. Your company research, question prep, and resume review all happen at the same time instead of three separate sessions.
When to use what
| Situation | Approach |
|---|---|
| Quick question, short task | Single agent (just ask Claude) |
| Important decision or fact-check | Dual verify (Claude + Gemini) |
| Big project with distinct parts | Agent team (parallel workers) |
Start with single agent for everything. Graduate to dual verify for important things. Use agent teams when a project has clear, independent parts that can be worked on simultaneously.
Chapter 9 The Upkeep Loop
Lobster season in Belize isn't just about catching lobster. It's the result of a year-round reef that's been maintained so it keeps producing. Your AI workspace is the same. A little maintenance keeps it sharp.
Equipment calibration, license renewals, CME credits. Healthcare professionals maintain their tools and credentials on a cycle. Your AI workspace needs the same discipline, just less often.
The monthly check-in (15 minutes)
Once a month, open Claude Code and run through this:
Review my workspace and brain file. Tell me: 1. What's outdated and needs updating? 2. What rules in my brain file are stale? 3. Are there empty folders or abandoned files I should clean up? 4. What's working well that I should keep doing? 5. Suggest one improvement to my setup.
The correction habit (ongoing)
Every time Claude does something wrong or not the way you wanted, add a rule to your brain file. This is the most important upkeep habit. Each correction makes every future session better.
Example corrections that become rules:
- "Always ask which platform before writing a post" (because it kept defaulting to Instagram)
- "Never use the word 'leverage' in my content" (because it sounds corporate)
- "When I say 'draft,' give me the complete final version, not an outline"
Just open your CLAUDE.md, add the rule to the Rules section, and save. That's it.
Monthly research refresh
Once a month, spend 30 minutes finding new content in your space:
- Use Gemini CLI to find new YouTube videos and articles on your topic
- Add them to your NotebookLM notebook
- Ask NotebookLM: "Based on my newest sources, what's changed since last month?"
- Update your workspace docs with anything new
This keeps your research library current and your content ideas fresh.
Weekly interview diary review
If you're actively interviewing, do a weekly review:
- Add any new recording transcripts to your workspace
- Add them to your NotebookLM notebook
- Ask NotebookLM: "Across all my practice sessions, what patterns do you see? Where am I improving? Where am I still weak?"
- Update your STAR stories and question prep based on findings
Certification tracking
Keep a simple tracker in your workspace. Tell Claude to remind you of upcoming deadlines:
Review my certification tracker at certifications/cme-tracker.md What's expiring in the next 90 days? What CME credits do I still need? Create a study plan if anything is urgent.
The quarterly question
Every three months, ask Claude this bigger question:
Look at my entire workspace. Be honest: - What am I not using that I should delete? - What's missing that would make this workspace better? - What would someone else need to know to use this workspace? - If you were starting fresh, what would you change about the structure?
Chapter 10 Quick Reference
Decision tree
I want to ask a quick question → Just type it into Claude Code
I want to research a topic deeply → Gemini CLI (discover) → NotebookLM (store + synthesize) → Claude (build)
I want to build something → Write a PRD first, use plan mode, then let Claude build it
I want to check an important answer → Dual verify (ask Claude + ask Gemini, compare)
I have a big project with multiple parts → Agent teams (parallel workers)
Something went wrong → Add a rule to your brain file so it never happens again
My workspace feels messy → Monthly check-in prompt (Chapter 9)
Key commands
| Action | Command |
|---|---|
| Start Claude Code | claude |
| Ask Gemini a question | gemini -p "your question" --yolo |
| Save your work (git) | git add . && git commit -m "message" && git push |
| Open NotebookLM | notebooklm.google.com |
| Create a new GitHub repo | gh repo create NAME --private --source=. --push |
| Open folder in VS Code | code . (from terminal) or File → Open Folder |
| Toggle terminal in VS Code | Ctrl + ` (backtick) or View → Terminal |
| Open extensions in VS Code | Cmd+Shift+X |
| Check Node.js is installed | node --version |
Essential prompts
| When You Need | Say This to Claude |
|---|---|
| Build your brain file | "Interview me and build a CLAUDE.md brain file" |
| Set up your workspace | "Set up folders for [describe areas] with brain files in each" |
| Plan before building | "Make a plan first. Show me what you'll create. Don't start until I approve." |
| Monthly review | "Review my workspace and brain file. What's outdated?" |
| Save and backup | "Commit my changes and push to GitHub" |
| Analyze a recording | "Analyze the transcript at [path]. What were my strong/weak points?" |
Official links
- Node.js (install this first)
- Claude Code Documentation
- VS Code (visual workspace)
- Gemini CLI on GitHub
- NotebookLM
- GitHub
- GitHub CLI
Pricing changes. I've linked to official pages rather than listing exact prices. Check the official sites for current pricing.