Full-stack and AI developer based in Hyderabad. I like building things that are small enough to finish and real enough to use.
Mostly TypeScript and Python — Next.js and Supabase on the front, FastAPI when there's a model involved. Currently interested in retrieval and in the parts of agent memory that turn out to be harder than they sound.
Arkaira — an online flower shop. Realtime stock that decrements when someone pays, Razorpay checkout, admin dashboard. Live here.
RecallDesk — a support copilot that remembers. Customer comes back saying "it's broken again", the agent already knows which ticket it was and what fixed it last time.
Anchor — for people coordinating care for an aging parent. Keeps the medication list, the lab results, the doctor's instructions in one place, so you're not reconstructing three months of history from memory in an appointment.
GitGlance — a CLI that scores a GitHub profile the way a selection committee would, and tells you what to fix. python glance.py <username>. Written because I wanted the feedback my own profile was getting, in a form I could argue with.
EnvGuard — typed environment variable validation for Node. One dev dependency, and it fails at boot listing every bad key instead of at 3am on a request.
SpamScope — Naive Bayes written out by hand, no sklearn, no numpy. I wanted to see the Laplace smoothing actually do something rather than take it on faith.
A batch of ML exercises I worked through — CNN on MNIST (99%), RAG chatbot with a browser UI, house price regression, movie recommender, iris classification. Useful for learning the shape of the problem. I keep them up because the code runs.
I'm open to hackathon teams and to interesting problems — the GitHub inbox works.