Anshum

..

[Anshum @TheAnshumDev]

I often find myself looking at people and quietly deciding who is struggling, who is happy, who deserves more, and who has less.

But who am I to judge?

Who am I to decide what someone truly needs? Who am I to tell God what He should give to someone else? I don't know the battles they have fought, the prayers they have whispered, or the peace they have found.

The only question that matters is this: Can I do something to help?

If the answer is yes, then I should help.

If the answer is no, then perhaps my opinion isn't needed.

Sometimes, compassion is more valuable than judgment, and humility is wiser than assuming we know someone else's story.

[Anshum @TheAnshumDev]

I have always thought that the hardest battles we fight are the ones no one sees, the ones inside our own minds. It is really easy to say we want to change but real self improvement is not about reading books or watching motivational videos. It's about sitting with yourself in the quiet, facing the parts of you that you would rather ignore. It's about owning up to your mistakes, forgiving yourself for them and deciding to do better even when no one is watching. Growth feels lonely sometimes like breaking apart the version of yourself you have clung to for years, but that is where it starts (letting go of who you were to make room for who you can be). In my opinion it is NEVER about perfection but about being honest with yourself and taking one small step forward every single day even when it hurts.

[Anshum @TheAnshumDev]

Most AI chatbots stop at:

User → LLM → Answer

But production AI needs much more.

User → Intent → Context → Query → Data → Intelligence → LLM → Answer

With fallbacks, validation, permissions, PII protection, and safety controls at every layer.

LLM for understanding.
Code for control.

Now I’m interested in how others are approaching this in production.

Do modern enterprise AI systems follow a similar hybrid architecture, or are agentic architectures replacing this kind of controlled pipeline?

[Anshum @TheAnshumDev]

Some Sundays are meant to be like this.

[Anshum @TheAnshumDev]

@abhinavspams Yep!!

[Anshum @TheAnshumDev]

The weekend is the only time we have to actually out-work the noise. Let's get it..

[Anshum @TheAnshumDev]

I know it didn’t always feel like you were moving forward, but you were. Every single day.
Life is just the sum of small decisions made in difficult moments. If someone has a better life, it’s because they made the better decision in those moments—over and over again.

[Anshum @TheAnshumDev]

Day 11 of the ML pivot.

While the world is hitting 'refresh' on their feed, I'm hitting 'run' on a training loop that finally makes sense.

The goal isn't to work 24/7.

The goal is to be so obsessed with the architecture of the future that "escaping" feels like a step backward.

[Anshum @TheAnshumDev]

The deeper I get into the math (Day 10 currently), the more I realize that the 'AI Revolution' is actually just a 'Fundamentals Revolution' in disguise.

[Anshum @TheAnshumDev]

The AI "Hype" cycle is a trap.
Don't spend your time: • Reading 100 threads on "Top 10 AI Tools."
Spend your time: • Mastering the Linear Algebra. • Understanding the Gradient Descent.
One group is building a feed. The other group is building a future.
Day 9. Signal > Noise.

[Anshum @TheAnshumDev]

Speed is for the first 6 months.

Moats are built on what you can do when the API hits its limit.
I’m spending Day 9 deep in the math of ML because I’m not building a wrapper—I’m building for the long game.

The "shortcut" era of AI is ending. The "Architecture" era is beginning.

[Anshum @TheAnshumDev]

@JashSolanki06 @rajshamani Ji..

[Anshum @TheAnshumDev]

Most people see a wall of code.

I'm currently deep in the "Uncanny Valley" of my own ML model. It's 60% accurate, 100% frustrating, and the only thing I've thought about for 48 hours.

There's a specific kind of madness in trying to teach a machine how to think...

[Anshum @TheAnshumDev]

1 hour of deep-diving into Gradient Descent > 10 hours of watching "AI Hype" videos. 📉

Focus is the only competitive advantage left.

Back to work. ⚔️

[Anshum @TheAnshumDev]

Most "AI Founders" right now are just building fancy wrappers for OpenAI's API.

The real winners are the ones currently suffering through Calculus and Linear Algebra. If you can’t explain the loss function, you don’t own the product—the API owners own you.

[Anshum @TheAnshumDev]

The "Junior Developer" is officially dead.

If your only skill is writing React components that an LLM can generate in 3 seconds, you aren't a developer—you're a prompter.

The only way to survive the next 24 months is to understand the math, the models, and the architecture.

[Anshum @TheAnshumDev]

Building apps is one thing. Understanding the "Brain" behind them is another.
I’ve decided to go back to the basics. Shifting my focus entirely to deep-diving into Machine Learning and AI architecture.
The goal: Move from "API Wrapper" to "Model Architect."

[Anshum @TheAnshumDev]

When I was a kid, I once asked my uncle what job he does.

He said, "I work in tech."

I told him, "Wow, that's such a nice job... you just sit in the office with a laptop all day and relax."

Today I finally understand - every profession has its own battles.

[Anshum @TheAnshumDev]

The more I learn ML, the more I realize:

One model can be powerful. But many models working together can be even better.

Today I explored ensemble methods:

Bagging, Boosting, Stacking, Random Forest, Gradient Boosting, XGBoost, AdaBoost.