I finished my CS degree at Guru Nanak Dev Engineering College in 2025. The first code I shipped to production was during a 2024 internship at Qspider — a MERN event platform with Stripe webhooks, JWT, and role-based access. It worked the way an exam answer works: complete, untested by anything weird.

The weird stuff arrived at Masters' Union — building the LMS for an MBA-equivalent EdTech program — mostly on the proctoring system. The first time the server CPU pinned at 100% with five users on it, I learned that an architecture decision can be wrong — not slow, wrong. The rewrite that followed (client-worker inference, SFU routing, OpenRouter-powered session summaries) was a lesson in the difference between optimisation and reconsideration.

The messaging refactor, the analytics pipeline, the event classifier — all variants of the same lesson: the gap between passes the demo and survives 200 concurrent users is where the engineering actually lives. Most of what I now believe about systems came from that year.

What I find most interesting in 2026 is LLMs as infrastructure problems. Not the chat UI — the queueing, the rate limits, the failure modes when inference takes eight seconds and the user is still typing. I want to work on the systems that hold that up. If you're building there, I'd like to hear about it.