DEV Community: Gowtham Jayavarapu

I Thought I Could Build It in 9 Days I Was Wrong

Gowtham Jayavarapu — Fri, 21 Aug 2026 07:12:09 +0000

9 days.
7 Codex accounts.
Multiple free credits exhausted.
Replit credits gone.
API limits.
An old laptop with 8GB RAM and basically no storage left.
Barely any sleep.
And somehow… I still kept building.

I started Hack Hydra, a 9-day HydraDB hackathon, thinking I had enough time and resources to build something ambitious.

Spoiler: I absolutely did not. 😅

I started with Codex and burned through the free credits across all 7 of my accounts.

Then I moved to Replit and used those credits too.

At that point, my backend was around 80% done.

Frontend?

Still pending.

Overall?

Maybe 40% of the project was actually complete.

And I had basically no credits left anywhere.

Then I found Freebuff, which gave me 6 free sections every day. That genuinely helped me keep moving.

But then came the API limits.

And this is where I started realizing how badly I had planned this.

Claude kept warning me:

“You’re over-scoping.”

“This probably isn’t buildable in the available time.”

And honestly, my mindset was basically:

“Fuck man, I know. Your job is to follow my commands, not tell me what’s possible.”

I thought this was vibe coding.

I thought if I could describe what I wanted, AI would just build it.

And building the code itself actually wasn't the hardest part.

Debugging was.

I’d say almost 70% of my time went into debugging.

Checking logs.

Reading errors.

Figuring out why an API wasn't working.

Testing the backend manually.

Fixing one thing.

Breaking another.

Running it again.

Then there was the design work.

Then connecting everything together.

Then testing again.

A feature that sounds like:

«“This should be easy.”»

can turn into:

«“Why the fuck is this returning an error?”»

for the next few hours.

And I kept adding more.

More features.

More complexity.

More things the project “should” have.

I wasn't just building a project anymore.

I was building something that was way too complicated for the time, credits, hardware and resources I actually had.

Then my laptop started fighting me too.

It's 5 years old.

8GB RAM.

Storage completely full.

Running the project and manually testing the backend on it was painful.

And somehow I kept going.

Then the final 2 days arrived.

And almost nothing was working.

No proper demo video.

Still debugging.

Still fixing.

Still trying.

One day left.

Still nothing.

Then the final 12 hours.

Barely any sleep.

At that point, I finally stopped trying to convince myself that everything would magically come together.

I started accepting the situation.

I went to sleep.

Woke up.

Opened my laptop.

And after 9 days of hard work…

my app was broken.

That hurt.

I couldn't even make the demo video.

I didn't finish what I imagined on Day 1.

But after accepting that, something interesting happened.

I started looking for the next hackathon.

And while doing that, I found an idea for my next project.

That's when I realized this 9-day journey actually gave me something much more useful than another finished project.

I learned how I need to build.

Next time:

Plan properly before starting.

Not just “I have 9 days, let's build.”

I need to understand what is actually possible with the time, hardware, APIs and resources I have.

Get the core project working early.

The goal should be to have something working well before the deadline, not to still be building the foundation on the final day.

Start the demo video when I'm around 90% done.

Not at the last minute when everything is on fire.

Add scope one thing at a time.

Get one feature working.

Then ask: Do I actually have enough time for another?

If yes, add it.

If not, ship what already works.

Stop treating AI like magic.

Vibe coding makes building faster.

It doesn't remove debugging.

It doesn't remove architecture.

It doesn't remove testing.

It doesn't remove design.

And it definitely doesn't remove time constraints.

I've done a 50-day buildathon.

I've done week-long hackathons.

I've done 3 months of 11Labs hackathons.

But this 9-day hackathon taught me something different.

Sometimes the biggest lesson doesn't come from successfully shipping.

Sometimes it comes from spending 9 days finding out exactly why you couldn't ship.

I got cooked by this hackathon. 🔥

But I also learned a lot.

And now I have an idea for the next one.

Hopefully this time…

I actually plan before I start cooking. 😭

Why Is the Normal Distribution So Important in Machine Learning and Statistics?

Gowtham Jayavarapu — Fri, 10 Jul 2026 10:11:48 +0000

If you've started learning machine learning or statistics, you've probably encountered the term normal distribution countless times. Whether you're reading about data preprocessing, hypothesis testing, feature scaling, or outlier detection, the normal distribution seems to appear everywhere.

At first, it can feel like everyone is obsessed with a simple bell-shaped curve. But there are good reasons for its popularity.

In this article, we'll explore what the normal distribution is, why it's so important, and how it influences machine learning and data science.

What Is a Normal Distribution?

A normal distribution, also known as a Gaussian distribution, is a probability distribution that forms a symmetrical bell-shaped curve.

It has three key characteristics:

Most values are concentrated around the center (the mean).
Values become less common as they move farther from the center.
The left and right sides are mirror images of each other.

A classic example is the distribution of adult heights. Most people have average heights, while very short and very tall individuals are relatively rare.

The same pattern often appears in measurement errors, biological characteristics, and many naturally occurring phenomena.

Why Does It Appear So Often?

The reason lies in the way the real world works.

Many real-world observations are influenced by numerous small, independent factors. Consider a person's height. It depends on genetics, nutrition, health, environment, and many other variables. Each factor contributes a little, and together they produce a distribution that is often close to normal.

Another powerful reason is one of the most important results in statistics: the Central Limit Theorem.

It states that if you repeatedly take random samples from almost any population and calculate their averages, those averages will tend to follow a normal distribution as the sample size grows.

This remarkable theorem explains why the normal distribution appears so frequently, even when the original data is not normally distributed.

Why Is It Important in Statistics?

The normal distribution makes statistical analysis much simpler.

Many statistical methods were developed under the assumption that data—or more commonly, the errors in a model—follow a normal distribution.

These include:

Confidence intervals
Hypothesis testing
t-tests
ANOVA
Linear regression assumptions

When these assumptions are reasonably satisfied, the mathematical results are reliable and easier to interpret.

Why Does Machine Learning Care?

Machine learning algorithms learn patterns from data, and the characteristics of that data directly affect their performance.

Some algorithms work better when features are approximately normally distributed because optimization becomes smoother and the model can learn more efficiently.

Examples include:

Linear Regression
Logistic Regression
Gaussian Naive Bayes
Linear Discriminant Analysis

It's important to note that not every algorithm requires normally distributed data.

Tree-based methods such as Decision Trees, Random Forests, and XGBoost are generally much less sensitive to the distribution of input features.

The Connection Between Normal Distribution and Outliers

Whenever people discuss normal distributions, they also talk about outliers.

An outlier is a value that is significantly different from the rest of the dataset.

Imagine the following salaries:

$40,000
$42,000
$41,000
$43,000
$2,000,000

The last value is an outlier.

Outliers can distort the mean, increase the standard deviation, and negatively impact many machine learning algorithms.

In a perfectly normal distribution, extremely large or small values are very rare. Therefore, detecting values that fall unusually far from the mean is a common method for identifying potential outliers.

Why Is Scaling Often Mentioned Alongside Normal Distribution?

Another concept that frequently appears is feature scaling.

Suppose you're building a model using two features:

Age: 20–60
Salary: 20,000–2,000,000

Because salary has much larger numerical values, algorithms that rely on distances or gradients may pay far more attention to salary than age.

Scaling transforms features so they have comparable ranges, allowing the algorithm to treat each feature more fairly.

While scaling does not make data normally distributed, many preprocessing pipelines first reduce skewness (if necessary) and then scale the data.

When Is Normal Distribution Not Necessary?

One common misconception is that every dataset must follow a normal distribution.

This is not true.

Many real-world datasets are naturally skewed.

Examples include:

House prices
Income
Social media followers
Website traffic

These datasets often do not resemble a bell curve, and that's perfectly acceptable.

Modern machine learning includes many algorithms that perform exceptionally well on non-normal data.

Instead of forcing every dataset to become normal, data scientists focus on understanding the data and choosing techniques that match its characteristics.

Key Takeaways

The normal distribution is important because it appears naturally in many phenomena and provides the foundation for a large portion of statistics.

Understanding it helps you:

Interpret data more effectively.
Detect unusual observations (outliers).
Apply statistical methods correctly.
Understand why some machine learning algorithms perform better under certain conditions.
Build stronger intuition about how real-world data behaves.

Perhaps the biggest lesson is this:

The normal distribution isn't important because every dataset follows it. It's important because it gives us a mathematical framework for understanding uncertainty, variability, and the behavior of data.

Once you understand the normal distribution, concepts like standard deviation, z-scores, confidence intervals, hypothesis testing, outlier detection, and feature scaling become much easier to grasp. It serves as one of the foundational building blocks of statistics, data science, and machine learning.