I work across the stack, but the thread running through most of it is the same: take something that currently needs a person babysitting it, and make it not need that anymore. Lately that's meant RAG pipelines on AWS Bedrock, an AST-based tool that writes its own failing tests before it fixes anything, and a voice app that schedules a task before you've finished saying it out loud.

• Research Intern, Deep Learning & Computer Vision — IIT (BHU) Varanasi
• Summer Research Intern, Training & Placement Cell — NIT Kurukshetra
• Software Engineering Intern, Full-Stack — Myschord (QuizerAI)

quelltest — open-source Python CLI that reads your code as an AST, finds the guard clauses nobody tested, and writes a test that fails before it writes the fix that makes it pass. No LLM key required for the core engine. 11,000+ downloads, 16 releases, no paid tier.

Lekha — meeting audio in, searchable transcript and structured minutes out, with a cloned voice for playback. Speech-to-text with speaker diarisation and automatic action-item extraction.

Bhagavad Gita Platform — certification platform built for ISKCON, with chapter-gated quizzes over an 18-chapter ingestion pipeline and an ontology-driven classifier that cut Bedrock token costs by roughly a third.

AlertyAI — say a task out loud, get it scheduled in under two seconds. Flutter and Kotlin on Android, Next.js on web, Sarvam AI underneath.

A couple of longer notes from the build log, if that's your thing:

• The engineering student who got tired of waiting to build something real
• I promised myself I wouldn't go home this summer

Open to internships, freelance work, and open-source collaboration — particularly anything that mixes applied ML with real product constraints.