DEV Community: Angkon-Dutta-Joy

The 80/20 Guide to Prompt Engineering (5 Tips That Actually Matter)

Angkon-Dutta-Joy — Mon, 09 Mar 2026 06:18:56 +0000

Prompt engineering is getting a lot of attention lately. But if you read most guides online, they often make it sound more complicated than it needs to be. We often see long frameworks, complex terminology, and endless lists of techniques. But in reality, we do not need twenty tricks to get good results from AI tools like ChatGPT or Claude. A few small habits create most of the improvement.

Think of it as the 80/20 rule. A small set of prompt techniques produces most of the results.

Here are five that matter the most.

1. Start with a clear request

The most common mistake people make is being vague. When the request is unclear, the response will usually be generic.

Instead of writing something like:

Tell me about marketing

Try giving the model more direction:

Explain five beginner digital marketing strategies for a small online clothing store

Now the AI understands:

the topic
the number of ideas
the context

Clear prompts lead to clearer answers.

2. Tell the AI how the answer should look

AI models do a much better job when you specify the format you want.If we do not mention a format, the model has to guess.

we can guide the structure with simple instructions like:

Answer in 5 bullet points
Explain step by step
Write a short paragraph
Return the result as a table

This small change often makes the response easier to read and much more usable.

3. Show one example

When we want a specific tone or pattern, giving one example works surprisingly well. For instance, imagine we want the AI to rewrite messages in a more professional way.

Rewrite professionally.

Example
Input: hey send the report
Output: Please send the report when available.

Now rewrite
Input: send me the file asap

With just one example, the model quickly understands the style you want.

we do not need many examples. Often one is enough.

4. Give the AI a role

Another simple trick is assigning a role before asking the question.

This helps the model respond from a more focused perspective.

Examples:

You are a senior Python developer. Debug this code.

You are a startup advisor. Suggest product ideas for solo founders.

This small bit of context can noticeably improve the relevance of the response.

5. Add constraints

Limits often produce better answers.

Without constraints, the model tends to generate long explanations that may not be very focused.

we can guide it with small boundaries such as:

Explain in under 100 words
Give 3 ideas only
Write for beginners
Avoid technical jargon

These limits keep the response tight and useful.

A simple prompt template

If you want a structure that works for many tasks, try this:

Role:
Task:
Context:
Constraints:
Output format:

Example:

Role: Startup advisor
Task: Generate SaaS ideas
Context: AI tools for small businesses
Constraints: 5 ideas, each under 20 words
Output: simple list

This structure keeps prompts clear without overthinking the wording.

Final thoughts

Prompt engineering does not have to be complicated.

Most of the time, better results come from doing a few simple things consistently.

Be clear about what you want
Describe the format you expect
Show an example when needed
Give the AI a role
Add reasonable limits

These habits work across different AI tools and models.

At the end of the day, writing a good prompt is really just clear communication.

And clear communication almost always leads to better answers.

A Deep Dive into Large Language Models (LLMs): Unpacking the Magic of AI Text Generation!

Angkon-Dutta-Joy — Mon, 31 Mar 2025 08:42:05 +0000

Hey everyone,

Ever wonder how AI like ChatGPT can chat with us, answer questions, or even write poems? 🤖 It might feel like magic, but it’s actually the result of some seriously cool technology called Large Language Models (LLMs). They’re the engines behind this AI conversation, and today, we’re going to break down what they are, how they work, and why they’re such a game-changer for the world of tech.

So grab your coffee ☕, sit back, and let’s take a fun (but informative) journey into the heart of AI!

What Exactly Are Large Language Models?

At their core, LLMs are AI systems built to understand and generate human-like text. They can do a ton of things—translate languages, write essays, summarize articles, answer questions, and even help with coding. Think of them like super-intelligent text generators that keep getting better the more they learn.

These models are usually built on something called the transformer architecture. That’s the secret sauce behind how they process text so quickly and accurately. Pretty neat, right? Let’s dive into how these transformers actually work.

How Do Transformers Work?

So, transformers are like the brainpower behind LLMs, and they’re pretty cool because they’re way faster than older models. You see, in the past, models used to process text word by word (like reading a book slowly, one page at a time). But transformers? They can process multiple words in parallel, making them way quicker at handling long sentences or paragraphs.

The magic ingredient here is something called self-attention. In simpler terms, this means the model doesn’t just look at each word in isolation. It looks at all the words in a sentence and figures out which ones matter most for the task at hand. So, it’s kind of like how you can read a sentence and instantly know which words carry the most meaning.

Training a Transformer – It’s Like Teaching AI to Read

Imagine trying to teach an AI how to write a novel or debug code. That’s pretty much what training a transformer is like. It’s a long process where the model learns to predict what comes next in a sentence, based on everything it’s read so far. Over time, it gets really good at understanding patterns and generating responses that sound natural.

After training, you can “fine-tune” these models for specific tasks, like answering questions or writing specific types of text. It’s like taking a generalist who knows a little about everything and turning them into a coding expert or a poet. 🎭

A Quick History of Transformer Models – From GPT-1 to GPT-4

Here’s a quick timeline of how transformer-based models have evolved:

GPT-1 (2018): The first real breakthrough in large-scale text generation. It was pretty basic, but it opened the door to what was possible.
GPT-2 (2019): The bigger sibling, GPT-2 had 1.5 billion parameters (a huge leap!), and it could generate much more coherent text. Plus, it started to show signs of zero-shot learning—the ability to perform tasks without having seen specific examples.
GPT-3 (2020): The AI that made everyone sit up and take notice, with 175 billion parameters. It could write anything from essays to code and even handle complex instructions.
GPT-4 (2023): This one’s a powerhouse. It can handle multimodal inputs—meaning it can process not just text, but images, audio, and video! Imagine an AI that can write a story, draw a picture, and make a video all in one go. Pretty cool, huh?

Fine-Tuning: Making LLMs Smarter (and More Useful)

Once a model has been trained on a huge dataset, you can “fine-tune” it for specific tasks. This is where things get fun. You can teach it to be an expert at something—like helping you write better code, follow instructions, or even have more natural conversations.

There are two main ways of fine-tuning:

Supervised Fine-Tuning (SFT): This is like giving the model examples and saying, “Here’s how I want you to answer this.”
Reinforcement Learning from Human Feedback (RLHF): This one’s a bit more advanced. It’s like playing a game where the AI gets rewards based on how well it follows your instructions. Over time, it learns to align with what you want, like being more helpful, truthful, or safe.

Prompt Engineering: How to Talk to Your AI

If you’ve ever played around with ChatGPT or similar models, you know that the way you ask a question can totally change the response. That’s where prompt engineering comes in. It’s like crafting the perfect question to get the best answer.

Some of the tricks include:

Zero-shot prompting: Just give the AI a direct instruction and let it figure it out.
Few-shot prompting: Give it a few examples to help it understand the task better.
Chain-of-thought prompting: This is where you guide the model step-by-step through the reasoning process, which helps it handle complex problems.

Mastering prompt engineering is like becoming a wizard who can summon just the right answers from your AI. ✨

LLMs in Action: What Can They Actually Do?

LLMs aren’t just for chatting—they can do a lot more! Here are a few things they’re amazing at:

Code generation: Need to write or debug code? LLMs can help you with that.
Text summarization: Got a long article or paper to read? LLMs can give you a concise summary.
Question answering: Ask it anything, and it’ll give you a detailed answer based on what it knows.
Creative writing: Want a poem or a short story? LLMs can create that for you too.
Translation: Need to translate a document? LLMs handle that effortlessly. #What’s Next for LLMs? As LLMs continue to evolve, we’re in for even more exciting developments. Imagine AI that understands and generates not just text, but images, videos, and even audio. The possibilities are endless! 🤩

Plus, with more companies releasing open-source models, the future of AI is looking incredibly collaborative and exciting.

Wrapping It Up

Large Language Models are more than just a buzzword—they’re a fundamental part of how AI is transforming the world around us. From writing code to creating art, these models are becoming more powerful and accessible every day.

The journey of LLMs is far from over, and as these models keep improving, the things we can do with them will only keep getting more incredible. So, stay curious, keep experimenting, and who knows? Maybe the next great AI breakthrough will come from you!

This version gives off a more casual vibe while still keeping the information clear and informative. It feels more like a conversation and a bit more engaging for your audience!

Fun Fact

I generated this blog post using an LLM itself! 🤯 All I did was provide it with my resources, and it wrote the whole post by itself! Isn’t that amazing? The power of AI is mind-blowing, and seeing it work like this really shows how far we've come. From just a few inputs to a fully written, coherent blog post—talk about next-level technology! 🚀

Cloud Computing Made Easy: An Introduction to Microsoft Azure Fundamentals

Angkon-Dutta-Joy — Mon, 15 May 2023 16:02:22 +0000

Microsoft Azure is one of the most popular cloud computing platforms, providing a wide range of services to help businesses and organizations meet their computing needs. To understand Microsoft Azure, it is important to first grasp the basic concepts of cloud computing.

In Microsoft Azure Fundamentals: Describe cloud concepts , there are 3 learning paths [Describe cloud computing, Describe the benefits of using cloud services and Describe cloud service types] which helps us to know about the concept thoroughly. This blog will cover the summary I learnt while completing the training path. Now I will share my learning outcome with you guys.

Before starting lets know about Azure.

What is Azure?

→ Microsoft Azure is a cloud computing platform with an ever-expanding set of services to help you build solutions to meet your business goals. Azure's web services allows businesses to host their presence in the cloud, as well as fully virtualized computers to manage custom software solutions. Additionally, Azure offers various cloud-based services such as remote storage, database hosting, and centralized account management. Furthermore, Azure provides new capabilities such as artificial intelligence and Internet of Things services.

What is Azure Fundamentals?

→Azure Fundamentals is a program consisting of three learning paths that introduce users to Azure's services and features. It includes interactive exercises and a temporary Azure portal environment for practice. No technical IT experience is required, but general IT knowledge is helpful. You should take this course if you:
● Have general interest in Azure or in the cloud
● Want to earn official certification from Microsoft (AZ-900)

AZ-900 Domain Area Weight

Describe cloud concepts 25-30%
Describe Azure architecture and services 35-40%
Describe Azure management and governance 30-35%

Cloud computing

#Introduction to cloud computing:-

→The module provides an introduction to cloud computing, covering general cloud concepts such as shared responsibility, different cloud models, and pricing methods. The learning objectives include defining cloud computing, describing the shared responsibility model, identifying appropriate use cases for public, private, and hybrid cloud models, and comparing cloud pricing models. This module may be a review for those already familiar with cloud
computing.

#What is cloud computing?
→ Cloud computing is the delivery of computing services, such as virtual machines, storage, databases, networking, IoT, ML, and AI, over the internet. Unlike traditional data centers, cloud computing is not constrained by physical infrastructure. This allows for rapid expansion of IT infrastructure without having to wait for the construction of a new data center. Like in cloud
computing services, users just have to pay for the resources he use. He doesn't have to think about the computational power or storage capacity and the hardware part is also taken care of
by the service provider. So the user doesn't have to worry about that.

#Describe the shared responsibility model

→ In this responsibility model physical security, power, cooling, and network connectivity are the responsibility of the cloud provider where the consumer is responsible for the data and
information stored in the cloud and also responsible for access security, meaning you only give access to those who need it.
The shared responsibility model refers to the distribution of responsibilities between the cloud provider and the consumer. This distribution of responsibilities depends on the type of cloud service:
Infrastructure as a service (IaaS), platform as a service (PaaS), and software as a service (SaaS).

In IaaS, the provider takes care of physical security, power, and connectivity, while the consumer is responsible for the rest, including operating system, applications, and data.
In SaaS, the provider takes care of almost everything, including applications, data, and infrastructure, with the consumer being responsible for only a small portion of the service.

In PaaS, the responsibility is evenly distributed between the provider and the consumer. The provider takes care of the infrastructure and middleware, while the consumer is responsible for the applications and data.
Understanding the shared responsibility model is important as it helps organizations determine the level of control and security they have over their data and infrastructure in the cloud.

The picture explains that organizations are always responsible for the data stored in the cloud,devices that connect to the cloud, and accounts and identities within the organization. The cloud
provider is responsible for the physical datacenter, network, and hosts. The specific responsibilities for operating systems, network controls, applications, identity, and infrastructure
depend on the chosen service model (IaaS, PaaS, SaaS).

Cloud Modules

The cloud models define the deployment type of cloud resources. The three main cloud models are: private, public, and hybrid.

Private cloud: A cloud used by a single entity, either hosted on site or in a dedicated datacenter offsite. Provides more control and security, but also comes with greater cost and fewer benefits than public cloud deployment.

Public cloud: Built, controlled, and maintained by a third-party cloud provider. Offers general public availability to anyone who wants to purchase cloud services.

Hybrid cloud: A computing environment that uses both public and private clouds in an interconnected environment. Offers flexibility and an extra layer of security by allowing users to
choose which services to keep in public or private cloud infrastructure.

There are other modules also like :-
Multi-cloud: A scenario where multiple public cloud providers are used to manage resources and security across different cloud environments.

Azure Arc: A set of technologies that helps manage your cloud environment, regardless of whether it's public or private cloud, hybrid, or multi-cloud.

Azure VMware Solution: A solution that allows VMware workloads to be run in Azure with seamless integration and scalability for those who want to migrate from a private cloud environment to public or hybrid cloud.

consumption-based mode

When comparing IT infrastructure models, there are two types of expenses to consider:

Capital expenditure (CapEx): A one-time, up-front expenditure to purchase or secure tangible resources (e.g. building a datacenter).
Operational expenditure (OpEx):Spending money on services or products over time (e.g.renting a convention center).
Cloud computing falls under OpEx because it operates on a consumption-based model.

Instead of paying for physical infrastructure, electricity, and security, you pay only for the IT resources you use. This consumption-based model has many benefits, including:

No upfront costs.
No need to purchase and manage costly infrastructure that users might not use to its fullest potential.
The ability to pay for more resources when they're needed.
The ability to stop paying for resources that are no longer needed.

With a traditional datacenter, estimating future resource needs is crucial. Overestimating leads to wasted money, while underestimating can lead to decreased performance. In a cloud-based model, you don’t have to worry about getting the resource needs just right.

Cloud computing allows you to rent compute power and storage from someone else’s datacenter, treating cloud resources like your own. Unlike in your own datacenter, you only pay for what you use, and the cloud provider takes care of maintaining the underlying infrastructure for you.
In summary, cloud computing is a cost-effective, flexible, and scalable solution for businesses of all sizes.

Describe the benefits of using cloud services

High Availability:
● High availability ensures maximum availability of IT resources, even during disruptions or events that may occur.

● Azure provides highly available cloud environments with uptime guarantees depending on the service.

● Service-level agreements (SLAs) outline these guarantees.

Scalability:
● Scalability refers to the ability to adjust resources to meet demand.

● Cloud resources are scalable, meaning you can add or subtract resources to handle increased or decreased demand.

● Scaling comes in two varieties: vertical and horizontal scaling.

● Vertical scaling increases or decreases the capabilities of resources, while horizontal scaling adds or subtracts the number of resources.

Reliability:
● Ability of a system to recover from failures and continue functioning.

● Azure's decentralized design supports a reliable and resilient infrastructure.

● Applications can be designed to automatically take advantage of global scale for
increased reliability.

Predictability:
● Predictability in the cloud lets you move forward with confidence.

● Performance and cost predictability are influenced by the Azure Well-Architected Framework.

Performance:
● Autoscaling, load balancing, and high availability support performance predictability.

● Autoscaling can deploy additional resources to meet demand, while load balancing helps redirect traffic to less stressed areas.

Cost:
● Cloud analytics and tools like TCO and Pricing Calculator can help predict future costs.

● You can adjust your resources as needed to better manage and predict cloud spend.

Governance:
● Set templates help ensure that all your deployed resources meet corporate standards and government regulatory requirements.

● Cloud-based auditing helps flag any resource that’s out of compliance with your corporate standards and provides mitigation strategies.

● Depending on your operating model, software patches and updates may also automatically be applied.

Security:
● You can find a cloud solution that matches your security needs.

● Infrastructure as a service provides you with physical resources but lets you manage the operating systems and installed software, including patches and maintenance.

● Platform as a service or software as a service deployments may be the best cloud strategies for you if you want patches and maintenance taken care of automatically.

● Cloud providers are typically well suited to handle things like distributed denial of service (DDoS) attacks, making your network more robust and secure.

Manageability:
Cloud computing offers two types of manageability options:

management of the cloud and
management in the cloud.

Management of the cloud involves managing your cloud resources, which can be done in several ways in the cloud environment. You can scale resource deployment automatically based on need, deploy resources using preconfigured templates instead of manual configuration, monitor resource health and automatically replace failing resources, and receive automatic alerts based on configured metrics, giving you real-time performance updates.
Management in the cloud involves managing your cloud environment and resources through various interfaces, including a web portal, command line interface, APIs, and PowerShell. These
options provide flexibility in managing your resources and allow you to choose the best method for your needs.

cloud service types

Infrastructure as a Service (IaaS),
Platform as a Service (PaaS), and
Software as a Service

(SaaS) are three different cloud computing models that offer various levels of management and control over IT resources.

IaaS is most flexible and provides physical resources such as servers, storage, and networking, which customers can manage themselves, including the operating system, middleware, and
applications.

PaaS offers a higher level of abstraction by providing customers with a platform to build and deploy applications without having to manage the underlying infrastructure. The customer only
needs to manage the application and data.

SaaS is the most abstract model and provides customers with access to software applications that are fully managed by the provider. Customers can use the software over the internet without having to install, manage, or update it.

How to make your python code faster

Angkon-Dutta-Joy — Tue, 28 Mar 2023 18:34:21 +0000

Python is a popular language with simple syntax and a wealth of libraries that is widely used in competitive programming. However, when it comes to solving complex algorithmic problems, many programmers struggle with the issue of Time Limit Exceeded. While it's true that Python isn't the fastest language out there, there are several techniques you can use to write efficient code and ensure that it gets accepted.

Use the appropriate data structure:

Using the correct data structure can have a significant impact on runtime. While Python has built-in data structures such as lists, tuples, sets, and dictionaries, many people tend to use lists in all cases. However, it's essential to use the right data structure depending on your task. For example, iterating over a tuple is often faster than iterating over a list.

Reduce the use of for loops:

As for loops are dynamic in Python, they take more time than while loops. Therefore, it's better to use while loops instead of for loops.

Use list comprehension:

List comprehension is a powerful tool that you should use instead of other techniques whenever possible. For example, if you need to list all the numbers between 1 and 1000 that are multiples of 3, list comprehension is a more efficient way to achieve this than using the append method.
Instead of

L = []
for i in range (1, 1000):
    if i%3 == 0:
        L.append (i)

Use -

L=[i for i in range(1,1000) if i%3==0]

Use multiple assignments:

Using multiple assignments can save time by assigning variables together in a single line instead of one by one.

#instead of 
a = 1
b = 2
c = 3
d = 4
#write
a,b,c,d=1,2,3,4

Avoid using global variables:

Global variables take more time to operate than local variables. Therefore, it's better not to use global variables unless necessary.

Use library functions:

Whenever possible, use library functions instead of writing your functions manually. Library functions are highly efficient and often impossible to match in terms of coding efficiency.

Concatenate strings with join():

While you can concatenate strings with the + operation in Python, it's better to use the join() method as it's faster. The + operator creates a new string and copies the old content at each step, while join() doesn't work that way.

#instead of 
concatenated_String = "Python " + "is " + "fun."

#write
concatenated_String = " ".join (["Python", "is", "fun."])

Use generators:

If you have a large amount of data in your list and need to use only one data at a time and for once, then use generators as they can save you time.

Avoid using dot operations:

Using dot operations can be time-consuming, as they first call getattribute() or getattr() and then use dictionary operations.

#instead of 
import math
val = math.sqrt(60)

#write
from math import sqrt
val = sqrt(60)

Use 1 for infinity loops:

Using while 1 instead of while True can reduce runtime.

While 1:
    ......

Try a different approach:

Sometimes, taking a different approach to writing code can lead to better performance. For example, instead of using nested if statements, you can use a single if statement with logical operators.

#instead of 
if a_condition:
    if another_condition:
        do_something
else:
    raise exception
#write
if (not a_condition) or (not another_condition):
    raise exception
do_something

Use sys for faster input:

Use the code below to get faster input time

import sys
input= sys.stdin.readline

Use speed-up applications:

To decrease runtime, you can use projects like Pypy and Numba. These applications can reduce the runtime of your program.

Use special libraries to process large datasets:

Python may not be the fastest language for processing large datasets, but you can use packages and modules written in C/C++, such as Numpy, Scipy, and Pandas, to optimize your code.

Use the latest release of Python:

Python is regularly updated and optimized, with each new release being faster than the previous one. Therefore, it's important to always use the latest version of Python.

These are just a few ways to speed up your Python code. There are several other techniques that you can use, so be sure to use a search engine to find them and write efficient code!

Python (Data types, Values and Variables)

Angkon-Dutta-Joy — Tue, 21 Mar 2023 13:00:37 +0000

Data Types

Data types are like the different categories of data that exist in Python - think of them as different "flavors" of data! There are five main groups of data types in Python, but there are also many other types available. First let's know about them. Don't worry if you don't understand a single term. we will learn about them deeply in the upcoming vlogs.

The first group is numeric type, which includes integers (positive or negative whole numbers without a fractional part) and floating-point numbers (real numbers with a decimal point). Think of them like different kinds of numbers on a number line!

a = 5  #(here 5 is an integer we stored in a continer named a which will be stored in our memory)
b = 12.65 #(it is a float data type. any fractional number will be float data type.)

Next up is boolean type, which includes just two values: True and False. This type is used for logical operations, like checking if something is true or false.

a=False
b=True

The third group is sequence type, which is an ordered collection of similar or different data types. Strings are a type of sequence made up of ordered characters - think of them like a bunch of letters in a particular order! Lists are also a type of sequence, but they can have elements with more than one data type. Tuples are another type of sequence, but we'll cover those later on.

a='python'
b="let's learn python"
c=[1,2,3,4,5]
d=['I','Love','Doughnut']

#here a and b is string data type and c and d is list data type

The fourth group is mapping type, which includes dictionaries. These are collections of key-value pairs, like a dictionary where each word has a definition.

thisdict = {
  "brand": "Ford",
  "model": "Mustang",
  "year": 1964
}

Finally, there's NoneType, which represents a null or no value at all. It's a special data type with a single value: None.

a = None (None have to start with capital letter N)

Data type checking

The type() function is commonly used for debugging and code correction purposes as it helps identify the type of an object or argument passed as a parameter. By returning the type of the object, the function provides useful information about the data being used in the code, which can be used to diagnose and fix errors.
Code

type("Hello World") 
type(110) 
type(123.456)
type(True) 
type(None)

Output

str
int
float
bool
NoneType

print() Function and it's use

The print() function is employed to display the value of an expression enclosed in parentheses (a pair of opening and closing brackets). It is typically used to generate output on the screen when writing a program, unlike in a command-line interpreter where it may not be necessary. The print() function is essential for observing and analyzing text output generated by a set of code. Further information regarding functions in Python will be discussed at a later time.
code

print('Hello World')

Output

Hello World

Variable and value

Variables are containers for storing data values.
Suppose you need to store a value 5 for further calculation. So, where will you store it? In computer memory right? But the name for that memory address will be too long for you to remember. So, why not we give it a name?

let call that container that will store our 5 in it is gpa.
So, we have to write,
gpa=5

we right the variable name in the left side and value on the right.

lets see how variable works in a real code.

check = 123
print(check)
print(type(check))
print("======")
check = "Hi"
print(check)
print(type(check))

Output

123
<class 'int'>
======
Hi
<class 'str'>

We can change the value of same variable with another value.

color='Red'
color='Green'
print(color)

Green

see, here color was red but then we wrote green for which the value changed and green is stored, but red lost.

Variable naming conventions

While naming the variables we need to follow certain rules and restrictions. These are called “Variable
naming conventions”. By following these our codes become more understandable by the Human
Readers (us). Few significant rules have been mentioned below.
Variable names-
● can have letters (A-Z and a-z), digits(0-9), and underscores ().
● should maintain snake_casing. That means each word should be separated by underscores()
● cannot begin with digits
● cannot have whitespace and special signs (e.g.: +, -, !, @, $, #, %.)
● are case sensitive meaning ABC, Abc, abc are three different variables.
● have to be meaningful with the data stored in it.
● cannot be too lengthy and too general, need to have a balance
● should not be written with single alphabets. Minimum length should be 3.
● cannot be a reserved keyword for Python. There are a total of 35 keywords.they are keywords that you can't use as variable name which are :

That's all for today....keep reading, keep learning. Also don't forget to practice.