DEV Community: Sandeep Maddheshiya

I Built a Visa Photo Tool After Realizing How Complicated “Just a Passport Photo” Can Be

Sandeep Maddheshiya — Thu, 13 Aug 2026 06:15:22 +0000

A few weeks ago, I was working on a small side project and noticed something surprisingly annoying.

Visa and passport photo requirements are not as simple as:

“Upload a passport photo.”

Different countries have different requirements.

Size, aspect ratio, background, head position, face visibility, file size… and sometimes the information is buried across government websites and PDF documents.

I kept thinking:

Why isn't there a simple way to check this before submitting the photo?

So I started building one.

That eventually became VisaReadyNow.

The interesting part wasn't actually building the image-processing logic. That was probably the easier part.

The harder part was figuring out what the requirements actually are.

I started going through official government and immigration sources country by country, comparing requirements, checking measurements, and trying to avoid making assumptions.

I also learned pretty quickly that “close enough” isn't good enough for something like this.

A 35×45 mm requirement isn't the same thing as a 35×35 mm requirement just because both are passport-photo sizes.

That pushed me toward a different approach:

Don't just generate a photo.

Explain why the photo passes or fails.

So now I'm working on making the tool more transparent — showing the requirements, linking back to the sources, and making it clear when something was last reviewed.

It's still a work in progress, but building it has taught me quite a bit about combining image processing, web development, SEO, and research.

You can see what I've been building here: (https://www.visareadynow.com/)

I'm curious about what's a small problem you've worked on that turned out to be much more complicated than you initially expected?

How I Reduced Website Latency by Optimizing the Right Things

Sandeep Maddheshiya — Thu, 13 Aug 2026 06:07:04 +0000

I recently spent some time optimizing one of my websites, and honestly, the biggest improvements didn’t come from some fancy optimization trick.

It came from measuring everything first.

I started looking at the actual request timings:

DNS lookup
TCP connection
TLS handshake
Time to First Byte (TTFB)
Total response time

Then I worked through the bottlenecks one by one.

Some of the things that helped:

• Reduced unnecessary server-side work
• Improved caching
• Optimized API requests
• Kept database queries lean
• Reduced unnecessary network round trips
• Tuned the VPS/server configuration
• Used CDN/edge caching where it actually made sense
• Removed things from the critical path that didn't need to be there

One thing I learned: a website can feel “fast” locally while being noticeably slower for someone sitting thousands of kilometers away.

So I started testing from different locations instead of relying only on my own browser.

The goal wasn't to chase a perfect Lighthouse score.

It was much simpler:

Make the first request reach the user as quickly as possible.

And the interesting part is that small improvements across several layers added up to a pretty significant difference.

Still experimenting with it, but this was a good reminder that performance optimization is mostly about finding the actual bottleneck instead of guessing.

What’s the biggest latency bottleneck you've found in one of your projects?

I Thought AI Interior Design Was Mostly a Gimmick Until I Built One

Sandeep Maddheshiya — Sun, 02 Aug 2026 19:04:07 +0000

For the longest time, I assumed AI interior design tools were just another "upload a photo and get a fancy render" demo.

Then I tried building one.

Turns out, generating the image is the easy part.

The difficult part is making the result look like something a person could actually imagine doing to their room.

One thing I noticed while testing is that people rarely ask for a complete redesign.

Nobody uploads a picture of their bedroom expecting AI to magically create a million-dollar renovation.

Instead, they ask things like:

"What if I paint this wall dark green?"
"Would wooden flooring make this room warmer?"
"Can I make this look more minimal without replacing everything?"
"How would this room look if it had better lighting?"

They're looking for reassurance more than inspiration.

That changed how I approached the project.

Initially, I spent most of my time trying to improve image quality. Better prompts, better models, better outputs.

But after watching people use it, I realized image quality wasn't the biggest problem.

Speed was.

People wanted to try ten different ideas in five minutes instead of spending hours jumping between Pinterest boards, Instagram posts, and random Google Images.

That felt much more useful than generating one perfect-looking room.

Another thing I underestimated was how personal interior design is.

Two people can upload almost identical rooms and ask for completely different styles.

One wants something cozy.

The other wants something modern.

Neither answer is objectively correct.

The AI isn't replacing creativity—it just helps people explore ideas faster.

That's probably the biggest lesson I took away from building this.

For anyone curious, the project is called ZyoraAI. I originally started it because I wanted to see how far current vision models could go with room redesigns, but it ended up teaching me more about user behavior than AI itself.

I'm still experimenting with it, and honestly, watching people interact with the product has been far more interesting than writing the code.