The gap worth solving
Interview prep for Indian engineering students is scattered across YouTube playlists, PDF dumps, and generic platforms built for a different market. Most of it tells you what to study. Almost none of it tells you how you are actually doing.
I kept watching smart friends walk into interviews underprepared, not for lack of effort but for lack of feedback. That was the gap: practice that talks back. Not another question bank, but a loop where every answer gets a response you can act on.
I decided to treat it the way I treat every product problem: find the gap, build the smallest thing that closes it, ship it, and let real usage tell me what is next.
Designing the practice loop
The core of PrepPilot is a tight loop: pick a track, answer questions the way you would in a real interview, and get immediate AI feedback on both the content and the gaps in your reasoning. The AI is not the product. The loop is the product; the AI just makes the loop fast enough to be worth repeating daily.
Every design decision followed from one rule: reduce the distance between practising and knowing where you stand. That killed a lot of tempting features early. No forums, no streak gimmicks, no content library for its own sake. Just practice, feedback, and progress you can see.
Owning the whole stack
I built PrepPilot the same way I build everything: the full stack, alone, with boring technology where possible. A modern React front end, an API layer that treats the AI calls as just another dependency to be cached and rate-limited, and authentication and payments wired in from the start because a product with paying users cannot bolt those on later.
The AI integration taught me the most. Prompting for consistent, structured feedback is an engineering problem, not a writing problem: schemas, retries, evaluation of outputs, and graceful degradation when the model gives you something useless. The pipeline work I do at Robomations showed up here constantly.
# the loop, in one trace POST /api/session/answer ✓ answer stored 12ms ✓ feedback generated 1.8s ✓ score + gaps returned → user sees exactly what to fix
Shipping, and the first rupee
PrepPilot went from first sketch to a deployed product because I refused to let it stay a side project. Deploy early, deploy ugly, fix what real users actually hit. The first paying user changed how I look at everything I build: a payment is the only piece of feedback that cannot be polite.
Since launch the roadmap is driven by usage, not by my guesses. The features that survived are the ones people came back for. Everything else got cut, and the product is better for it.
