PracticalX Toolkit — Day 3 · Teaching AI Your World

What this toolkit is: A practical companion to Day 3 of Learning AI Out Loud. No theory. No fluff. Prompts and exercises you can use today — whether you're an everyday AI user or someone building AI systems professionally.

💡

Foundation

The Core Idea

AI was trained on the world but knows nothing about your world. RAG — Retrieval Augmented Generation — is the concept of giving AI your specific context so its responses stop being generic and start being genuinely useful. Think of it as filling a filing cabinet your AI can search before every conversation.

AI trained on the world is impressive.
AI trained on your world is useful.

🟢

For everyday users — no technical knowledge needed

Try It Now

Any AI tool — ChatGPT, Claude, Microsoft Copilot

10 minutes

Responses that feel personal, not generic

EXERCISE 01

Build Your Personal Context Brief

This is the single most impactful thing you can do to improve your AI experience today. Copy this prompt and fill in the brackets:

Copy · Paste · Fill in the brackets

"I want you to act as my personal assistant. Here is the context you need to know about me: I work as [your role] in [type of organization]. My current priorities are [list two or three things you are working on]. The language and terms we commonly use are [list any specific terms or acronyms]. My preferences when you help me are [bullet points / plain English / concise answers]. Please use this context in everything you help me with."

What to notice: Ask AI something you've asked before — without context first, then with context. Compare the two responses. The difference will be immediate.

EXERCISE 02

Fix a Generic Response

Next time AI gives you an answer that feels like it could apply to anyone, try this:

Copy · Paste · Fill in the brackets

"The response you just gave me is too generic. Here is specific context about my situation: [describe your specific world, role, organization type, current challenge]. Now rewrite your response using this context."

EXERCISE 03

Plan Something Personal

Try planning something — a project, a presentation, a trip, an event. First ask AI without any context. Then repeat the same request but start with:

Copy · Paste · Fill in the brackets

"Before you respond, here is everything you need to know about my specific situation: [paste your context — who you are, your preferences, your constraints, your goals]. Now help me with: [your request]."

What to notice: The second response will feel like it was written for you rather than for anyone.

🎯

Your Challenge For Today

Write your personal context brief — who you are, what you do, your current priorities, your preferences. Save it somewhere accessible. Paste it at the start of your next AI conversation. Think of it as the onboarding document you'd give a brilliant new colleague on their first day.

🔵

For technical users — implementation and architecture focused

Go Deeper

Familiarity with APIs and LLMs helpful

30 minutes to explore

A practical RAG implementation starting point

Understanding RAG Architecture — Three Stages

Retrieval

When a query comes in, the system searches your knowledge base for the most relevant content using semantic similarity rather than keyword matching.

Augmentation

The retrieved content is injected into the prompt as additional context before the query reaches the language model.

Generation

The model responds using both its training knowledge and the retrieved context from your specific knowledge base.

EXERCISE 01

Evaluate Your Knowledge Sources

Copy · Paste · Fill in the brackets

"I want to implement a RAG architecture for my organization. Our key knowledge sources include: [list your documents, wikis, SharePoint, Confluence, databases]. For each source, help me evaluate: data freshness requirements, appropriate chunking strategy, and retrieval priority. Then suggest a phased implementation approach starting with the highest value source."

EXERCISE 02

Vector Database Selection

Copy · Paste · Fill in the brackets

"Compare Pinecone, Weaviate, Chroma, and pgvector for an enterprise RAG implementation. Our specific requirements are: [scale requirements, latency expectations, security constraints, existing infrastructure]. Recommend the best fit and explain the key trade-offs for each option."

EXERCISE 03

Build an Evaluation Framework

Copy · Paste · Fill in the brackets

"Help me design a comprehensive evaluation framework for a RAG implementation. I need to measure retrieval accuracy, response relevance, hallucination rate, and end-to-end latency. Suggest specific metrics, a testing methodology, and a baseline benchmarking approach I can use before and after implementation."

EXERCISE 04

Chunking Strategy

Copy · Paste · Fill in the brackets

"I have the following types of documents in my knowledge base: [describe your document types — policies, reports, FAQs, technical docs]. For each document type, recommend the optimal chunking strategy — chunk size, overlap, and any metadata I should attach to each chunk to improve retrieval accuracy."

🎯

Your Challenge For Today

Map your organization's top five knowledge sources. For each one, score it on two dimensions — how frequently the content changes, and how much it would improve AI output quality if included as context. That two-by-two matrix is your RAG implementation priority list.

⚠️

A Note on Privacy and Security

Safe to include: Your role, your priorities, your working preferences, general organizational context
Check your policies first: Internal documents, strategic plans, customer data
Never include: Personally identifiable information, financial data, anything governed by compliance requirements
When in doubt, use general descriptions rather than specific data

Resources Worth Exploring

Search "RAG explained" for current plain English explainers
LangChain documentation — the most widely used RAG framework
LlamaIndex — strong for document-heavy RAG implementations
Your AI tool's documentation on custom instructions and memory features