Key contributions:- Worked extensively on unsupervised clustering techniques to meet various use-cases.- Worked on NLP using LLMs and embedding techniques.- Conceptualized and developed the backend for a UI-based ETL pipeline called Data Model supporting connectors for multiple data stores. Data Model has the ability to query database(s) from one or more sources, apply transformations like merging data across databases, aggregation, etc. Contributed in designing the UI for the same. It is now one of the core modules being used across products.- Contributed significantly to enhancement of the ML pipeline to streamline the processes.- Developed various ML models to address use-cases for timeseries forecasting.- Dockerized all ML-services to streamline development and production deployment.- Studied, benchmarked and implemented TimescaleDB as a database solution for the company's ML products.

BS-MS (Dual Degree) majoring in PhysicsArea of Specialization - AstrophysicsMS Thesis - CMB Component Reconstruction and Angular Power Spectrum Estimation using Neural Networks

I used the 1.2 Jy IRAS Redshift Survey data to obtain best-fit parameter values for the galaxy selection function using Maximum Likelihood Estimation (MLE) and MCMC.I implemented convergence diagnostics for MCMC by checking the initial Markov chain’s auto-correlation as a function of lag. I further improved the MCMC estimation using Cholesky decomposition of the covariance matrix of accepted parameter values chosen in a way to minimize correlation. This led to manyfold improvements of our results.I pursued this project during the Coronavirus lockdown.

I simulated polymer folding using the bond fluctuation model. I also computed the mean first-passage looping times for various polymer configurations including free-chain and grafted-chain polymers.This helps in comparing the looping times of various polymer configurations which further enables polymer based research.

The project was funded by the INSPIRE mentorship.I worked on ‘21cmSense’ Python package to get sensitivity estimates of radio telescopes to the HI 21cm Power Spectrum. I studied the package, upgraded its dependencies, reduced its run-time, and added a new function to help faster execution. I got introduced to the basics of 21cm cosmology and interferometry.The work done in this project was part of a larger effort to derive a ML-based approach to detect the HI 21cm signal. Sensitivities estimated here were used to train neural networks to detect the signal and aid analysis.

The project was funded by the INSPIRE mentorship.I implemented the MCMC (Markov Chain Monte Carlo) Metropolis-Hastings algorithm to obtain best-fit values of the density parameters of the universe using SNe data. I applied least-squares fitting using the Levenberg-Marquardt algorithm and numerical integration.Finding accurate best-fit values of the Density Parameters of our universe improves our understanding of the current universe and its evolution. It also aids further research on early universe and cosmology.