AI can feel like a foreign language — full of confusing buzzwords.

I kept running into these terms in docs and videos without fully understanding them, so I decided to slow down and make sense of the most common ones.

Here’s my simple breakdown.

---

1️⃣ Prompt

A prompt is simply the message or instruction you give the AI.

A prompt is like talking to a very smart person who can do almost anything - but you need to give clear instructions.

✔️ Examples of prompts:

• “Summarize this text in simple language.”
• “Generate a list of frontend interview questions.”
• “Explain embeddings like I’m 10 years old.”
• “Find leave balance for this employee.”

✔️ A good prompt:

• sets context
• gives instructions
• gives examples (optional)

Prompts are becoming a real skill because: Better prompt → Better output → Better AI-powered apps

2️⃣ Tokens

A token is a small chunk of text that the model reads or generates. It’s not exactly a word — it might be:

• a word
• part of a word
• a punctuation mark

✔️ Examples:

• “frontend” → might be 1 token
• “developer” → might be 2 tokens
• “a”, “,”, “.” → each are tokens

✔️ Why tokens matter:

• AI models charge per token
• They can only “remember” a certain number of tokens (context window)
• Bigger answers = more tokens
• Longer documents = more tokens

Token awareness helps you understand the cost, speed, and memory limits of AI systems.

3️⃣ Context Window

The amount of information an AI model can “remember” at one time. Measured in tokens.

✔️ Example:

• GPT-4 Turbo: ~128k tokens
• Claude Opus: 200k tokens
• Some models: 1M tokens

✔️ Bigger context window = longer conversations + bigger documents.

⭐️ Quick Example to Make It Clear

If the context window is 100 tokens:

• your input uses 40 tokens
• the model output uses 60 tokens

If you exceed this, the model “forgets” earlier parts of the conversation.

4️⃣ LLM (Large Language Model)

I like to think of an LLM as a static brain. It’s trained on a huge amount of text, but only up to a certain point in time.

So:

• If something happened 5 minutes ago, the LLM won’t know.
• It also doesn’t automatically know your company’s internal data.
• And you can’t just “plug” it into your system and expect it to know everything.

✔️ In short:

An LLM is powerful, but limited to what it was trained on.

That’s why we need extra techniques to make it useful in real-world apps.

5️⃣ RAG (Retrieval-Augmented Generation)

This is where things get exciting. If an LLM is a static brain, RAG gives it access to your real data, almost like handing it a memory book.

✔️ How I understood it:

RAG lets you feed your own data to the LLM so it becomes personalized for your use case.

✔️ This means your AI system can finally:

• Answer questions about your company
• Search your documents
• Stay up to date
• Give accurate, customized responses

Without retraining the model.

6️⃣ Embeddings

It simply means converting text into vectors (numbers) based on meaning. Two words with similar meaning will have similar vectors.

✔️ Example:

• “Vacation” → vector A
• “Holiday” → vector B

These vectors will be close to each other even though the words are different. This is how AI understands similarity in meaning.

7️⃣ AI Agents

An AI Agent = LLM + memory + tools + ability to take actions. It’s not just a chatbot that replies — it can do things.

✔️ Example I understood well:

A company HR AI agent that can:

• read your company policies
• know your leave balance
• understand your request
• and actually apply for leave for you

✔️ Agents don’t just answer — they act.

8️⃣ Vector Database

A Vector DB works differently from traditional databases:

• It stores your text as vectors (embeddings).
• And retrieves data based on meaning, not exact words.

✔️ Traditional databases match exact words. So if you query:

SELECT * FROM leaves WHERE reason = 'vacation';

You’ll get results with “vacation” …but miss results with “holiday.”

✔️ But in Vector DB, searching “vacation” will also find “holiday”. ✔️ Super useful for semantic search, chatbots, RAG systems, etc.

9️⃣ LangChain

An abstraction layer that makes writing AI apps easier.

✔️ You can mix and match components like:

• models
• prompts
• vector DBs
• tools
• agents
• memory

…without rewriting your whole application every time. It’s like React for AI — reusable building blocks.

---

💡 My beginner insight:

AI isn’t magic. It’s math + data + smart engineering.

---

🌿 What’s Next?

Understanding these buzzwords removed a lot of friction for me.

Now, while reading documentation or watching AI videos, I can focus on how systems work instead of getting stuck on terminology.

If you’re also learning AI step by step, feel free to follow along. Let’s keep learning — one concept at a time 🌱🤝