Web Developerπ©π»βπ» | Technical Writer | JavaScript | React | Figma | Content Writer @softaims_ | DMs open
Remote
Joined July 2021
- Tweets9.7K
- Following429
- Followers13.9K
- Likes35.5K
Ishrat retweeted
Okayy deployed it today, come try it free! π₯
I built this primarily for my bullet journaling, but can be used to write anything.
It's completely local, whatever you write stays with you.
Live site: alicalimli.com/shiruya
Ishrat retweeted
The this keyword is one of those JavaScript concepts that looks simple but often causes confusion.
At its core, this refers to the context in which a function is called.
The important part is:
this is determined by how a function is called, not where it is defined.
Here are the common cases:
1. Inside an object method
const user = {
name: "Ishrat",
greet() {
console.log(this.name);
}
};
user.greet(); // Ishrat
Here, this refers to the user object.
2. Inside a regular function
In non-strict mode, this can refer to the global object.
In strict mode:
"use strict";
function greet() {
console.log(this);
}
greet(); // undefined
3. Inside an event handler
button.addEventListener("click", function () {
console.log(this);
});
Here, this refers to the element that triggered the event.
4. Arrow functions are different
Arrow functions don't have their own this.
They inherit this from their surrounding scope.
const user = {
name: "Ishrat",
greet: () => {
console.log(this.name);
}
};
user.greet(); // undefined
This is one reason you should be careful when using arrow functions as object methods.
The key takeaway
Don't memorize "this always means the current object."
Instead, ask:
"How was this function called?"
That usually tells you what this refers to.
Understanding this makes objects, event handlers, classes, callbacks, and frameworks like React much easier to reason about.
Ishrat retweeted
What is Fine-tuning? π
Fine-tuning is the process of continuing to train a pre-trained model on a new dataset.
Instead of simply providing the model with additional context like RAG,
you're updating the model's learned parameters.
In other words :
β RAG changes the context.
β Fine-tuning changes the model.
β¨ What does fine-tuning actually change?
A model contains millions or even billions of learned parameters from its original training.
Fine-tuning updates some (or all) of those parameters using your own dataset.
The goal isn't to add documents.
It's to change how the model behaves.
β¨ When should you use fine-tuning?
Fine-tuning is useful when you want the model to consistently :
β Follow a specific writing style
β Produce outputs in a particular format
β Learn a domain-specific task
β Improve performance on a repeated workflow
Think of it as teaching the model a new behavior, not giving it new documents.
β¨ When should you use RAG instead?
If your goal is to work with :
β Company documentation
β PDFs
β Knowledge bases
β Wikis
β Frequently changing information
RAG is usually the better choice.
The documents stay outside the model and are retrieved only when needed.
The important thing to remember :
Fine-tuning doesn't turn the model into a searchable knowledge base.
It teaches the model how to behave, not what documents to remember.
That's why AI products like ChatGPT, Claude, Cursor, GitHub Copilot, Perplexity, and Bug0 generally use RAG for dynamic knowledge and fine-tuning when they need to change the model's behavior or capabilities.
Ishrat retweeted
8 AI skills everyone should learn in 2026:
1. Context Engineering
AI is only as good as the context you give it.
Learn how to provide:
β’ Goals
β’ Examples
β’ Data
β’ Constraints
β’ Output format
Better context = better results.
2. Workflow Automation
Stop doing repetitive work.
Automate workflows like:
Lead β AI research β Email β CRM β Follow-up
Learn tools like:
β’ n8n
β’ Make
β’ Zapier
3. AI-Assisted Coding
You don't need to be a developer.
But you should know how to:
β’ Explain problems
β’ Review AI-generated code
β’ Debug errors
β’ Ship faster
4. RAG & Knowledge Systems
Teach AI to use your own data.
Build:
β’ Internal knowledge bases
β’ Customer support bots
β’ Company assistants
β’ Document search
5. AI Integrations
The real magic starts when AI can use your tools.
Learn:
β’ APIs
β’ Webhooks
β’ MCP
β’ Authentication
Connect AI to Notion, Slack, CRMs, calendars, databases, and more.
6. Multimodal Content Creation
Turn one idea into:
β’ Blog
β’ Thread
β’ Carousel
β’ Video
β’ Image
β’ Podcast
One idea. Multiple formats.
7. AI Agents
Move beyond simple prompts.
Learn how to build AI agents that can:
β’ Plan
β’ Reason
β’ Use tools
β’ Remember context
β’ Complete multi-step tasks
This is where AI is heading.
8. AI Search Visibility
People search on:
β’ Google
β’ ChatGPT
β’ Claude
β’ Gemini
β’ Perplexity
Learn SEO, AEO, and GEO.
Create content that's easy to understand, well-structured, and worth citing.
Don't try to learn everything.
Pick one skill.
Build a real project.
You'll learn more from shipping than watching another AI tutorial.
You don't need to master all 8 today.
Pick one skill.
Build a real project.
Keep improving.
That's how you actually learn AI.
---------------------------
Thanks for reading!
Follow @ishrratumar for deep dives into each of these skills. I'll be covering them one by one in upcoming posts.
Ishrat retweeted
What is RAG ? β¨
RAG stands for Retrieval-Augmented Generation.
It's a technique that allows an AI application to retrieve relevant information before asking the LLM to generate a response.
Instead of relying only on what the LLM learned during training, the application first looks up information from external sources, then includes it in the prompt.
How RAG works? β¨
1. You ask a question.
2. The application searches a knowledge source.
It might retrieve :
β Company documentation
β PDFs
β Wikis
β Database records
β Knowledge bases
β Vector databases
The retrieved information is added to the prompt as additional context.
The LLM generates a response using that information.
Isn't that just context? β¨
Not exactly.
Context is everything the LLM receives before generating a response.
RAG is one technique the application uses to retrieve relevant information and turn it into context.
The LLM doesn't know whether that information came from you or was retrieved automatically.
It simply receives the final context and generates the response.
A simple way to remember it :
π Context = Everything the LLM sees.
π RAG = One way the application builds that context.
As AI applications continue to evolve, RAG has become one of the core building blocks for working with company knowledge, documentation, and constantly changing data.
That's why you'll find it in many modern AI products, including ChatGPT, Claude Code, Cursor, and Bug0.
_____________________
Thanks for reading till the end!π
In tomorrow's post, I will explore more about RAG.
Ishrat retweeted
Where does AI memory actually live? β¨
An AI application is still just a software application.
The LLM is only one component inside it.
Just like traditional applications store user data in databases, AI applications store memory outside the LLM.
π Short-term memory
Short-term memory is used while the AI is working on a task.
It can include :
β Your current conversation
β Files you've uploaded
β The task you're working on
β Temporary task information
Once the task or conversation ends, this memory is usually discarded.
It is often stored in application state or a cache.
πΎ Long-term memory
Long-term memory persists across sessions.
It can store :
β Your preferences
β Ongoing projects
β Important facts you've asked the AI to remember
β Frequently used workflows
This information is retrieved whenever it's relevant to a future conversation.
Depending on the application, this information may be stored in :
β SQL or NoSQL databases
β Vector databases for semantic search
β Knowledge bases
β File systems or cloud storage
When needed, the application retrieves the relevant information and sends it to the LLM as additional context.
Can AI memory change over time?
Yes.
Modern AI systems can :
β Update memories
β Delete outdated information
β Retrieve only what's relevant
β Organize memories by user, project, or task
Some systems even support versioning or branching memories for different workflows.
A common misconception is that every LLM has its own memory.
In reality, products like Claude Code, Bug0, ChatGPT, Cursor and many other AI applications usually maintain a single application-managed memory that's shared across different models.
The key idea is simple :
The LLM does the reasoning.
The application manages the memory.
That's why memory in AI is as much a software engineering problem as it is an AI problem.
Ishrat retweeted
everything is gunna work out in the end
if it hasn't worked out then it is not yet the end
Ishrat retweeted
Most people start AI with models.
I did too.
Then I hit a quieter question that made everything clearer:
How does a computer store what it knows about the world?
ALT Knowledge Representation - AI
Ishrat retweeted
What happens when you die:
They divide up your shit.
They summarize your life in 500-1000 words.
People who knew you less say sorry to people who knew you more.
Everyone eats, drives home, and wakes up the next day and goes to work.
Whatever youβre worried about wonβt be in those 500 words.
You can dare greatly or not at all, but youβre gonna die either way.
Might as well squeeze every motherfucking drop out.
Ishrat retweeted
I built this thing called Clicky.
It's an AI teacher that lives as a buddy next to your cursor.
It can see your screen, talk to you, and even point at stuff, kinda like having a real teacher next to you.
I've been using it the past few days to learn Davinci Resolve, 10/10.
Ishrat retweeted
Many are asking if they can move their Next.js projects from Vercel to Replit?
Yes!
- Go to Import then GitHub
- enter repo URL
- Agent will take care of the rest
It will set up the dev and deployment environments!
For companies moving lots of work, happy to help + discount.
Ishrat retweeted
Hereβs how you can use famous AI tools to deploy your products faster:
1) Use Perplexity AI for research.
2) Use ChatGPT to turn that research into a clear plan.
3) Use v0 .dev to design your product
4) Use Cursor to write the code.
5) Use Claude to fix bugs.
All of this?
Costs about $100/month.
Ishrat retweeted
Claude 3.7 Sonnet + Replit Agent v2
It's been a HUGE week for AI-powered development. Two major releases just dropped that are changing how indie hackers & solo founders build products.
Here's my experience with both:
1/ Anthropic's Claude 3.7 Sonnet
Just tested this and I'm genuinely blown away. This isn't just another incremental AI update - it's redefining what's possible for developers.
What's new:
- The new "extended thinking mode" solves complex problems that stumped earlier models
- Generating code that works the first time
- Command-line interface that feels like having a senior dev on your team
The most impressive part? It doesn't just write code - it writes *good* code with proper architecture and clean patterns. You can create entire Next.js applications in one shot.
2/ Replit Agent v2
Replit also just launched their second-gen AI coding assistant, and it complements Claude perfectly.
The standout features:
β’ Build full applications from plain English commands
β’ Real-time design preview as you describe what you want
β’ An intuitive workflow for non-tech founders
At $15/month (Core plan), it's saving weeks of development time and thousands in contractor costs.
What this means for indie hackers/solo founders
The barrier to entry for building software just dropped dramatically. Ideas that would have taken months can now be prototyped in days.
MVPs that cost $20K+ can now be built for a fraction of that. Features that required deep technical expertise are suddenly accessible to non-coders.
We are seeing the advancement of real-time software development.
3/ My take
These tools aren't replacing developers - they're supercharging them. The real value is in the problems you solve and the products you design.
The winners will be the founders who learn to leverage these tools effectively while focusing on what truly matters: solving real customer problems.
People are building crazy stuff.
Need proof?
Check out what @SullyOmarr built with Claude 3.7 Sonnet using this simple prompt:
"please create me a saas landing page template designed extremely well. create all components required"
Ishrat retweeted
If youβre into health and have NoCode skills, this is a great opportunity
I'm starting a new side project and I need a CEO/co-founder to run it.
I won't say exactly what it is but:
- Health related
- Involves laboratory testing
- Riding a major trend
- Public good/public service angle, but can likely make profit
Email me if you're interested.
You:
- Early stage of your career, but you've done stuff before
- Ready to dive in immediately
- Know how to build websites using NoCode tools like Webflow
- Passionate about public health
- Willing to work for a low starting salary (weighted towards bonus and equity)
Ishrat retweeted
Your Conversion Rate is the outcome of 2 crucial steps:
1. A strategic approach to turning visitors into customers
2. Consistent optimization to maximize conversions at each stage
For the first step, you need to:
- Decide WHAT actions you want users to take
- Ensure your product truly solves their problems
- Make your value proposition crystal clear
Track these 3 key points in your SaaS pipeline.
First step is easier because YOU control the user experience.
But it gets trickier with the second step:
You need to turn potential into paying customers.
You need to:
- Go from "I want X% to convert"
- To "X% of my users are converting"
You don't decide anymore, YOUR USERS decide.
That's why the second step is harder and ongoing:
- Optimize your website and onboarding
- Use content to showcase value
- Stay consistent with your messaging
- Keep testing and improving (even when it's frustrating)
Keep refining your funnel to achieve the conversion rates you desire.
Remember:
- Average B2B SaaS trial-to-paid: 25-30%
- Rates vary widely (17.4%-51%) based on factors like:
β’ Industry and product complexity
β’ B2B vs B2C
β’ Marketing channels
β’ Subscription model
β’ Product maturity
Higher conversion rates = more customers = lower CAC