Blog

One of my favourite projects from college didn't start as anything ambitious. I just wanted to see if I could move my cursor around without touching anything — hold up a hand, point a finger, watch the pointer follow. It turned into a full virtual mouse built in Python, using a live camera feed to track hand position and gestures and translate them into cursor movement and clicks.

The interesting part wasn't the cursor moving — that's almost the easy bit. It was everything around it: getting consistent hand detection under different lighting, deciding which gesture should count as a "click" without it triggering by accident every time you moved your hand, and smoothing the motion so the cursor didn't jitter every time your hand shook slightly. A lot of what looks like "AI magic" from the outside turns out to be careful, unglamorous tuning once you're actually building it.

It's still one of the projects I bring up first when people ask what I've built, mostly because it's the one where you can actually see the result — wave your hand, and something on the screen listens.

Before my internship at TCiA Global, cloud computing was mostly theory to me — diagrams of regions and availability zones, and a vague sense that "the cloud" meant someone else's server. That changed fast once I was actually provisioning EC2 instances on AWS and virtual machines on Azure for real client workloads.

Working across both platforms side by side was, in hindsight, one of the more useful parts of the internship. You start noticing how differently AWS and Azure name the same idea, and how much of "cloud skills" is really about reading documentation quickly and not being afraid to break something in a test environment. I spent a good chunk of my time on the less glamorous work too — configuring backup failovers, setting up AWS SNS alerts, and writing SOPs and monitoring dashboards so reports didn't depend on someone remembering to check a console every morning.

I also ended up managing client websites, which wasn't originally what I expected from a "cloud computing" internship — but updating layouts and content for actual clients taught me just as much about working with requirements and deadlines as any AWS console did.

"Systems Engineer" can sound abstract until you're actually doing the job. In practice, my day-to-day at TCS is a mix of a few concrete things: patching applications to keep them current and secure, running performance tests to catch problems before users do, debugging issues that show up in production, and validating databases to make sure the data behind everything is actually correct.

None of that is particularly dramatic on its own, but it adds up to something I've come to appreciate: most of what keeps an application reliable happens quietly, before anything breaks. A patch applied on schedule is a problem that never becomes an incident. A performance test that catches a slow query is an outage that never happens. It's a different kind of satisfaction than shipping a flashy new feature, but it's satisfying all the same.

It's also taught me to read systems, not just write code — to look at an issue and figure out where in the stack it's actually coming from, rather than assuming the first plausible explanation is the right one.

Not everything worth building ends up as a headline project. Alongside the internships and coursework, a handful of smaller things stuck with me. At the Smart India Hackathon, my team built an AI-integrated smart yoga mat, and I worked on the Android app interface — posture detection and real-time feedback, built under the usual hackathon deadline pressure. At Techgium, the L&T-hosted hackathon, I helped analyze the architecture for an AI and sensor-based concept to help people with insomnia, and worked on the automation and app-based controls around it.

I also spent some weekends on a couple of machine learning models just to practice the full pipeline end to end — one predicting house prices from location, size and amenities, another assessing vehicle health from sensor data, both using regression and random forest. And on a more domestic note, I reconfigured my home network router to enable shared network storage, which sounds simple until you're the one troubleshooting why a device suddenly can't see the shared drive anymore.

That last one nudged me toward something I've been slowly picking up since: the basics of networking, and a bit of ethical hacking on the side — mostly out of curiosity about what systems look like from the other side of the fence. Not a specialty yet, but the kind of curiosity that tends to turn into a skill if you keep pulling the thread.