Projects
Production backend work from the last five years, research from IIT Gandhinagar before it, and a few things built because the problem was interesting.
A platform that collects global trends from TikTok, YouTube and Instagram through Apify and public data sources, then uses Gemini to turn the raw signal into structured insight for clients in beauty, cosmetics and travel. I built the backend services in Python and Node.js plus a data migration tool — 15% faster backend — and deployed the GraphQL and FastAPI microservices on Google Cloud Run behind CI/CD.
A fee-payment REST API spanning multiple gateways, letting administrators monitor transactions and reporting across more than 100K payments a month. I led the database and API work — Django REST Framework, Google Cloud Storage, microservices — and built 15+ custom modules with continuous integration across AWS Lambda and Google Cloud Functions.
A 12-layer neural network that returns the probability a chest X-ray shows COVID-19 infection — a fast preliminary screen before a formal test. Trained with clinicians on data from Ahmedabad hospitals and open sources, validated at 94.8% accuracy, and released as a public web tool tested by the Indian Institute of Public Health. I led it as principal investigator under Prof. Krishna Prasad Miyapuram.
My M.Tech thesis, published at Brain Informatics 2021. A 2D convolutional neural network that classifies Alzheimer’s disease from MR images, outperforming the state of the art on accuracy, sensitivity and specificity across all three brain planes. Built by preprocessing T1-weighted MRI volumes from the ADNI database into 94,536 sagittal, coronal and axial slices, then discarding the non-informative ones and correcting for head movement and noise.
Identified unusual user behaviour in large public WhatsApp groups by classifying message content into spam, advertisement, offensive content and everything else, then reading the pattern of who posts what.
An optimisation model built on particle swarm optimisation that maximises the distance between students sitting the same paper — the exam-hall version of “don’t put these two next to each other.” Written up on Medium.
A TCP-based, multi-client cloud storage system written in C, handling concurrent requests with secure access control. The server keeps an individual log file per user and enforces that nobody can read or write anyone else’s files. Adapted into an internal tool at Liftoff.
SVM, linear regression, naive Bayes, decision trees and random forests implemented in Python without a library doing the interesting part — the best way I know to actually understand an algorithm.
Uses the Google speech API on a two-minute audio recording to gauge how well schoolchildren speak, behind a REST API with Facebook, Google and custom sign-in.
My first real web project: a web-based alumni portal built for Tezpur University during a 2015 internship, backed by MySQL.
More code — Django experiments, a product-catalog REST service, stock analysis, TA material for MA-602 — lives on GitHub.