AI That Actually Matters: Real Problems, Real Impact

Fair warning, I’m a bit rusty. It’s been years since I wrote anything that wasn’t Code or a Corporate email. After a gap longer than I’d like to admit, something finally pulled me back to writing. Bear with me if the rust shows.

I’ve been thinking about this for a while now. Not the usual “AI will take our jobs” debate or the “ChatGPT wrote my email” kind of use case. The AI real world problems India is sitting on are much bigger, much more personal, and almost nobody in the industry is building for them.

There are real problems. Big ones. And the tools to fix them finally exist.

Let me explain what I mean.

A Bit of Background

If you grew up in India and tried to get into a decent engineering college, you know the drill. Percentages. Cutoffs. Eligibility criteria that seem to multiply every year.

I went through it. So did most of my batch. For years I just accepted it as the way things are. Working in tech and seeing what’s possible now, I started thinking: wait, a lot of this didn’t have to be this way.

Much like how good software architecture principles help us remove unnecessary complexity from code, AI can help remove unnecessary gatekeeping from real-world systems. The problem was never the rule. It was the scale at which nobody could enforce or evaluate things fairly.

The 33% vs 75% Problem: AI Real World Problems India Ignores

Here’s something that’s always bothered me, and I’ve never heard anyone actually say it out loud.

CBSE says you need 33% to pass Class 12. That’s the official rule. Score 33%, you’re done, you’ve passed, here’s your certificate, go live your life.

Now here’s where it gets interesting.

Back in our time (around 2006 onwards), if you wanted to appear for IIT JEE, you needed at least 60% in your Class 12 boards just to be eligible. Not to get in, just to sit for the exam. So if you scored 58%, the same board that gave you a pass certificate also made sure you couldn’t even walk into the IIT exam hall. This 60% criteria was the official JEE rule for general category students from 2006.

Fast forward to today, and the rule has changed but not in the way you’d hope. The 60% bar is gone. You can appear for JEE Main now without a minimum percentage. But even if you clear JEE Main, even if you clear JEE Advanced (which has roughly a 3% success rate and is considered one of the toughest exams in the world), you still can’t get admission to an IIT unless you have 75% in your boards or are in the top 20 percentile of your state board.

So let me get this straight. You can crack one of the hardest exams in the world and the door still doesn’t open because of your board percentage.

The board that said you passed at 33% is now the same yardstick being used to say you’re not good enough at 74%. Both rules exist. Side by side. Nobody has fixed this in twenty years.

The reason isn’t mysterious either. IITs get lakhs of applications. Even after JEE Main filters things down, around 2.5 lakh students qualify for JEE Advanced. The 75% rule is just one more way to shrink that pile. It’s not a quality measure, it’s a crowd control measure.

Where AI Actually Fixes This

The only reason these cutoffs exist is because evaluating lakhs of people properly, at scale, consistently, and fairly, is beyond what any human system can handle. It takes too long, costs too much, and produces wildly inconsistent results.

AI removes that excuse entirely.

If you can evaluate answer papers automatically, accurately, and consistently, you no longer need an arbitrary percentage to pre-filter people. You can let everyone sit. The exam itself becomes the filter, which is how it should have been all along.

A student who scored 40% can appear for JEE. Most won’t crack it. That’s fine, that’s what the exam is for. But at least their story doesn’t end on a technicality.

This is one of the most meaningful AI real world problems India could tackle, and yet almost no one in the startup ecosystem is talking about it seriously.

The Paper Checking Mess Nobody Talks About

Here’s a story from my engineering days I’ll never forget.

One of my friends, not the most academically inclined guy, was in a direct admission batch. In one particular exam, he told us he just wrote the question back as the answer. Every single question. The question was the answer. That’s it.

He scored 60 out of 70.

Now I can’t verify this. What makes it believable is that around the same time there were actual published news reports of young children being found checking university answer papers at some affiliated colleges. Not teachers. Kids.

Before you say “that’s an exception”, is it really? Teachers across India are dealing with lakhs of answer sheets every year, tight turnaround times, and very little support. Things slip. Quality of evaluation varies wildly. They’re human and the volume is simply not manageable.

Now imagine this instead. An AI agent reads each paper, evaluates the answer based on concept accuracy, not word count, not handwriting, and gives a score. Engineering answers should be precise. A two-mark answer should be two lines. If you nailed it in three words, you get full marks. The AI checks for that.

A second agent then challenges the first. Disagreements get flagged. A human reviewer only sees the flagged ones, maybe 5% of total papers. The rest go through clean, fast, and fair.

This isn’t some futuristic idea. It’s a multi-agent workflow. Anyone who’s played around with LLMs seriously in the last couple of years knows this is buildable right now.

The hiring problem, since we’re at it

Same story, different context.

A few years back I interviewed with a large Indian IT services company, one of the big ones. Their eligibility criteria included a minimum academic percentage (I believe it was around 80%) and also this gem: if you had a gap of two or more years in your education, you were automatically disqualified.

I had 10+ years of industry experience at the time. Shipped products. Managed teams. Done real work.

But apparently my gap year from God knows when still mattered more to their screening system than a decade of actual output.

And look, I get why companies do this. They receive thousands of applications. Nobody has time to actually read them all. So they build filters. The filters are blunt, unfair, and often have nothing to do with capability but they reduce the pile and that’s all the system cares about.

AI changes the math completely. If you can meaningfully screen a thousand applications in the time it used to take to screen fifty, you don’t need the blunt filter anymore. You can actually evaluate people.

Indian roads, I had to bring this up

I know this feels like a detour but stick with me.

Anyone who’s driven in India knows it’s a bit of a philosophical experience. Traffic lights are suggestions. Lane markings are decorative. Rules technically exist but enforcement is, let’s say, selective.

The reason is straightforward. You can’t put a traffic cop at every intersection in a country with 1.4 billion people. The math doesn’t work.

AI-powered cameras can. They don’t get tired. They don’t take chai breaks. They don’t look the other way because the car has a particular number plate. They just watch, record, and flag.

This is already rolling out in some cities. It works. And it’s the same pattern as everything else in this post. The problem was never the rule, it was the enforcement bottleneck. AI removes the bottleneck.

Why isn’t anyone building this?

Because it’s easier to sell an internal AI dashboard to a corporate client than to fix a broken national examination system.

One has a procurement team, a budget, and a quarterly review. The other requires government partnerships, years of trust-building, and the patience to wait a long time before you see results.

So we keep building slightly faster ways to summarise meetings and the real problems stay exactly where they are.

I’m not judging, I understand the economics. But I also think whoever’s willing to do the harder thing here is sitting on something genuinely significant.

Wrapping up

AI isn’t going to change the world by writing better marketing copy.

It’ll change the world when a kid in a small town who scored 68% gets to actually attempt an IIT exam. When a professional with twelve years of experience isn’t filtered out by a gap year they took in 2009. When an answer paper gets evaluated by something that actually reads it.

That’s the version worth building.

Got your own example of a broken system that AI could fix? Drop it in the comments. Would love to hear what you’ve seen out there.

Made it this far? Subscribe for more. Cheers!