Projects

Master thesis: trained UNet-family and RETFound models under differential privacy and benchmarked mechanisms that recover the accuracy lost to privacy noise.

Implemented custom Kubernetes scheduling policies and measured workload interference across co-located latency-critical and batch jobs.

Adapted state-of-the-art 3D instance segmentation to architectural scenes, scaling the class vocabulary 13.8x on ScanNet++.

Compared Opacus and fast differential privacy for training on medical images, measuring the speed, memory, and privacy trade-offs.

Implementation of examples across communication, databases, high performance computing and privacy.

Transformed data into insights with classifier evaluation, Lucene indexing, NLP, embeddings, RAG, and decision tree visualizations.

Compared gradient clipping techniques in DP-SGD and measured their impact on model accuracy.

Implemented algorithms with perceptron, classification, regression, boundary detection, image restoration, and deep neural network feature design.

Prediction of loan approval based on applicant details using logistic regression model.

A python package which uninstalls everything in your virtualenv enviornment

A free, open url scraper for Ali Express integrated with Google Spreadsheet helping drop-shipping business.

A Python package which makes very simple to run things in the background of a python application, be it a CLI app, or a web app

pip plant is similar to what is npm install in javascript

A simple perceptron implementation.

An URL-shortening service, similar to TinyURL.com or goo.gl, but with no persistent storage.

A simple game of Tic Tac Toe, built using python.

Shows top 5 traits shared between the two Twitter(X) accounts. For example: you and a celebrity of your choice.