Enter your search text in the box above

Select a result to preview

Assignment 6 Solve Questions

Q1. What is an API?

API stands for Application Programming Interface. In simple words, an API is a bridge that allows one software system to talk to another software system in a structured way.

We can think of it like a waiter in a restaurant. You do not go inside the kitchen and cook your own food. You give your order to the waiter, the waiter takes it to the kitchen, and then the kitchen sends the food back through the waiter. In the same way, when one app wants something from another app, it sends a request through an API, and the other system sends back a response.

For example, when a website shows a weather report, the website usually does not calculate the weather itself. It calls a weather API. When a payment app confirms a transaction, it often uses a banking API. When we use ChatGPT inside a project, we usually connect to the model through an API.

An API is important because it creates rules for communication. It defines:

Without APIs, every software system would need direct manual integration, which would be messy, insecure, and hard to scale.

So the real idea is this: an API is not the intelligence itself, not the database itself, and not the app itself. It is the communication contract between systems.

2. What is RAG?

RAG stands for Retrieval-Augmented Generation.

This means the AI does not answer only from what it learned during training. Before answering, it first retrieves useful information from an outside source such as notes, PDFs, company documents, websites, or a database. Then it uses that retrieved information to generate a better answer.

This is powerful because normal AI models have limited memory about your personal or company-specific data. They may know general things about programming, business, or history, but they do not automatically know your class notes, your private files, your research papers, or your company SOPs. RAG solves that problem by giving the model relevant context at the time of answering.

A simple flow of RAG is:

  1. User asks a question.
  2. System searches the knowledge base.
  3. Most relevant chunks of information are found.
  4. Those chunks are sent to the LLM.
  5. The LLM writes an answer using that context.

So RAG is not a separate AI brain. It is a system design pattern that improves an LLM by attaching retrieval to it.

The big benefit of RAG is that it helps the model become:

But RAG is only as good as the data quality, chunking quality, retrieval quality, and prompt quality behind it.

3. What is the difference between RAG and Agentic AI and Machine Learning?

These three things belong to different levels of the AI stack, so comparing them properly is very important.

RAG

RAG is a method for improving answers by retrieving relevant information before generation. It is mainly about giving better context to an LLM.

Agentic AI

Agentic AI is when AI does not just answer once, but works toward a goal. It may plan steps, use tools, call APIs, search documents, write code, check outputs, and decide what to do next. In other words, it behaves more like a worker handling a task than a chatbot giving a one-shot reply.

Machine Learning

Machine Learning is the broader field where systems learn patterns from data instead of relying only on fixed rules written by humans. It includes many techniques such as regression, classification, clustering, recommendation systems, and deep learning.

Core difference

Machine Learning is the broad learning field.

RAG is a retrieval architecture usually used with LLM systems.

Agentic AI is a goal-oriented behavior layer built on top of models and tools.

A simple analogy

So they are not competitors. They operate at different levels.

4. As a student, where can I use RAG?

As a student, RAG can be extremely useful because most of your important knowledge is not general internet knowledge. It lives inside your own notes, textbooks, PDFs, teacher handouts, lecture slides, past question papers, and personal summaries.

You can use RAG in many practical ways:

Study assistant from your own notes

You can upload your class notes and ask questions in simple language. Instead of getting a generic answer, the system can answer from your exact material.

Exam revision

You can ask:

Weak-topic detection

If your documents are uploaded properly, you can ask the system to identify which concepts are connected, where you are likely to get confused, and what you should revise first.

Assignment support

RAG can help you refer to your actual syllabus or source material while writing answers, instead of randomly generating unrelated content.

Career preparation

You can build a RAG system from:

Then you can ask focused questions from your own preparation material.

The biggest value for students is this: RAG turns scattered notes into a searchable brain.

5. As a non-technical person, explain how a RAG works behind the scenes.

Let us explain this very simply.

Imagine you have 500 pages of notes. You ask the system, "Explain recursion from my class material."

If the AI reads all 500 pages every single time, it will be too slow and expensive. So instead, the system prepares your data in advance.

Step 1. Your documents are collected

Your notes, PDFs, files, or web pages are uploaded into the system.

Step 2. The documents are broken into smaller parts

Instead of storing one giant note, the system splits the content into smaller chunks. For example, one paragraph, one section, or one concept at a time.

Step 3. Each chunk is converted into meaning form

The text is converted into embeddings. Embeddings are numerical representations of meaning. This allows the computer to compare ideas based on similarity, not just exact words.

Step 4. Those embeddings are stored

They are stored in a database, often a vector database.

Step 5. You ask a question

When you type a question, the question is also converted into an embedding.

Step 6. The system searches for similar chunks

It compares your question embedding with stored embeddings and finds the most relevant note sections.

Step 7. The relevant text is sent to the AI model

Now the model receives:

Step 8. The AI writes the final answer

The answer is generated using both the model's language ability and the retrieved knowledge from your notes.

So behind the scenes, RAG is doing two jobs:

That is why it is called Retrieval-Augmented Generation.

6. What is a database?

A database is a structured system used to store, organize, update, and retrieve data.

It is more than just a random file. A database is designed so that data can be managed properly, searched quickly, and used safely by applications.

For example:

If data were stored only in normal text files, handling large amounts of information would become slow and confusing. A database solves that by providing structure.

Common things a database helps with:

A database is important because modern software needs memory. If the frontend is the visible face of an app, then the database is one part of the app's memory system.

7. How should I use AI for decision making? Should I fully rely on AI for every task?

AI should support your decision making, not replace your responsibility.

This is one of the most important mindset lessons. AI is excellent at:

But AI is not truly responsible for the consequences of your life, career, money, health, or relationships. You are.

So the correct way to use AI for decision making is:

Use AI for clarity

Ask it to structure your thoughts, compare pros and cons, create decision frameworks, or show possible risks.

Use AI for research support

Ask it what factors you should consider before making a decision. This helps widen your thinking.

Use AI for simulation

You can ask:

But do not blindly obey it

AI can sound confident even when wrong. It may miss real-world context, emotions, personal values, legal realities, or recent information.

You should never fully rely on AI for every task because:

The best model is:

AI for support, human for judgment.

Especially in serious decisions, AI should act like a smart assistant, not like the final authority.

8. Why does AI make mistakes? What is human in the loop?

AI makes mistakes for many reasons.

It predicts, it does not truly understand like a human

Most generative AI systems are pattern prediction engines. They generate the most likely next sequence based on training and context. That means they can produce fluent language without true grounding.

Training data is incomplete

If the model has never seen enough good examples, or if the data had noise, bias, or gaps, the output can also be weak.

Context can be poor

If the prompt is vague, missing details, or misleading, the output can also become wrong.

Retrieval can fail

In RAG systems, sometimes the wrong chunks are retrieved. Then even a good model may answer from the wrong evidence.

Real world changes

AI may not know the latest laws, prices, events, or company-specific facts unless connected to fresh sources.

Overconfidence problem

One dangerous thing about AI is that it often presents wrong answers very confidently. This creates false trust.

What is human in the loop?

Human in the loop means a human is involved in checking, guiding, correcting, approving, or supervising the AI system.

This is important because AI should not always run independently in sensitive tasks.

Examples of human in the loop:

So human in the loop means AI assists, but a human still controls quality and final accountability.

This is one of the safest and smartest ways to use AI in the real world.

9. What is frontend and backend? In software why is backend needed? Why can we not do everything in frontend?

Frontend is the part of software that users can directly see and interact with.

Examples:

If you open a website and click, type, scroll, or view content, that visible experience is the frontend.

Backend is the hidden logic running behind the application.

It handles things like:

Why is backend needed?

Because frontend alone is not enough for real applications.

If you put everything in the frontend:

For example, if a payment rule, admin permission rule, or user verification rule is placed only in the frontend, a user may bypass it by modifying browser-side code.

Backend acts like the secure control room of the software. It decides what is allowed, talks to the database safely, protects secrets, and processes operations reliably.

Simple analogy

Customers see the counter, but the real control happens behind it.

10. What is the difference between database and backend?

Many beginners mix these two, but they are not the same thing.

Database

The database stores data.

Backend

The backend controls logic and operations.

The backend usually talks to the database, but it does many more things than just storage.

For example, suppose you have a notes app.

The database stores:

The backend handles:

So the database is like storage memory, while the backend is like the manager operating that memory.

The database does not usually decide business logic on its own. The backend sits in the middle between frontend and database and makes the system work properly.

11. As a non-tech person, explain: if I want to create a project where I upload my notes, how will frontend, backend, and database connect? How will the AI connect using API? If I want to create my own RAG, how will it work?

Let us walk through this as a complete project flow.

Suppose you want to build a website where students upload notes and then ask questions from those notes.

Step 1. Frontend

The frontend is the website interface. This is where the user:

So the frontend collects user actions and displays results.

Step 2. Backend

When the user uploads a file, the frontend sends that file to the backend.

The backend is responsible for:

So the backend is the main coordinator.

Step 3. Database

The database stores normal application data such as:

If you are using RAG, then you may also use a vector database to store embeddings for semantic search.

Step 4. AI API connection

Your backend talks to the AI provider using an API.

For example, when a user asks a question, the backend may send a request like:

The AI model returns the answer through the API, and the backend sends that answer back to the frontend.

Step 5. Full RAG flow

If you want your own RAG system, the full flow is:

  1. User uploads notes from frontend.
  2. Frontend sends files to backend.
  3. Backend extracts text from files.
  4. Backend breaks text into chunks.
  5. Backend generates embeddings for each chunk.
  6. Embeddings are stored in a vector database.
  7. User asks a question in chat.
  8. Backend converts the question into an embedding.
  9. System searches the vector database for similar chunks.
  10. Relevant chunks are collected.
  11. Backend sends question plus chunks to the LLM API.
  12. LLM generates the answer.
  13. Backend sends final answer to frontend.
  14. Frontend displays the answer to the user.

What is the main insight here?

Frontend shows.

Backend controls.

Database stores.

API connects.

LLM generates.

RAG grounds the answer using your own knowledge.

That is the full flow in simple language.

12. As a non-tech person, explain why I need to use a vector database over a normal database. What is a normal database and how does a vector database work? Give a simple example.

This is a very important concept for anyone building AI products.

What is a normal database?

A normal database stores structured information in a traditional way.

For example:

This is excellent for exact data lookup.

If you ask:

a normal database works very well.

But what is the limitation?

A normal database is not naturally built for meaning-based search.

Suppose your notes say:

"Large language models predict the next token based on context."

Now you search:

"How do AI models generate the next word?"

The wording is different, but the meaning is similar. A normal database mainly works well with exact matching or structured query rules. It may not understand that both sentences are conceptually related.

What is a vector database?

A vector database stores embeddings.

Embeddings are numerical representations of meaning. When text is converted into embeddings, similar ideas are placed closer together in a mathematical space.

So instead of searching by exact word match, a vector database searches by semantic similarity.

Simple example

Let us say your notes contain these three lines:

  1. "Neural networks learn patterns from data."
  2. "Photosynthesis happens in plants."
  3. "Deep learning uses many layered neural networks."

Now suppose you ask:

"Which parts of my notes talk about AI learning systems?"

A vector database can understand that sentence 1 and sentence 3 are related to your query, even if the exact words are different. It will retrieve those lines because their meaning is close.

Why use vector database in RAG?

Because RAG needs relevant chunks based on meaning, not just exact keyword match.

If a student asks:

"Explain how the model predicts output"

the system should still find notes that say:

"The language model generates the next token from prior context."

That is semantic similarity, and vector databases are designed for that.

Final difference

Normal database:

Vector database:

Best practice in real products

In many AI systems, both are used together.

So the real answer is not that vector databases replace normal databases. They solve a different problem. You use a vector database when you want the system to search by meaning instead of only exact text matching.