I'm a 3rd-year CSE student at Keshav Memorial, currently interning as a Security Engineering Intern at CyberParadigm. Most of my time there goes into agentic AI for offensive security I can't get into specifics of what we're building (NDA), but broadly: hardened attacker environments, and an AI agent that tries to reason its way through a pentest instead of following a fixed script. Turns out getting a model to know what it doesn't know is a much harder problem than getting it to do the task in the first place. Before this internship I built KrishVind, an AI agricultural platform that started as a Smart India Hackathon entry and somehow ended up in production 50+ REST APIs, live on Vercel, actual users. I didn't expect a hackathon project to outlive the hackathon. Also CTF-active. Placed 52nd out of 589 teams in HackTheBox's Global Cyber Skills Benchmark this year, which I'm still deciding whether to be proud of or annoyed I didn't crack the top 30. What I'm actually curious about: where LLMs are genuinely useful as reasoning agents for security work, and where they're just confidently wrong in a new way. Still figuring that line out.
Security Engineering Intern
CyberParadigm
An AI agricultural platform for small farmers: crop disease detection from a photo, real-time market prices, government scheme eligibility, weather-linked planning. It started as a Smart India Hackathon entry and never really stopped Found an exposed API key during a later audit and moved the AI calls server-side to fix it. Also added client-side OCR for document verification, so sensitive documents never actually leave the user’s device, a DPDP Act consideration I hadn’t thought about when I first built the feature
Official SIH problem statement (PS 26071) for the Ministry of Earth Sciences: forecasting flood risk by fusing satellite, radar, and weather-model data I lead the team and own the backend API and deployment infrastructure everyone else’s modules sit on top of Re-verified a teammate’s hydrology risk-engine module myself before merging it in: fault injection, swapping the DEM data to see what broke When a teammate became unavailable partway through, I took over their dashboard module and checked it against real pipeline output instead of the mock data it had been built against
B.Tech in Computer Science & Engineering
Keshav Memorial Engineering College
Google Cybersecurity Professional Certificate
Google (via Coursera)
Sept 2025Attendee · Oct 24, 2026