Biography
I currently work as a Data Scientist at Vayana Network, where I analyze and interpret company’s data to identify key patterns, trends and possible avenues for business growth strategies.
I am broadly interested in machine learning and its intersection with vision, language, healthcare, and other areas, with a focus on building inclusive and accessible language and vision models that can be applied across diverse domains and languages.
Interests
- Artificial Intelligence
- Machine Learning
- Natural Language Processing
- Computer Vision
Education
PG Diploma in Advanced Studies and Research (DipASR) in CS, 2022
Ashoka University, India
BSc (Hons) in CS, 2021
Ashoka University, India
Highschool Diploma, 2017
Budhanilkantha School, Nepal
Work Experience
Data Scientist
- Processing and managing the company’s data efficiently and securely by building a central data repository on AWS through data pipelines and internal libraries.
- Investigating the company’s business network through data analysis and graph modeling to identify key patterns, trends, and potential customers shaping the company’s business growth strategies.
Data Science Intern
- Designed the architecture for a fitness recommendation application based on wearable devices and led the team toward implementing the first milestone.
- Assessed the effectiveness of different anti-aging drugs and products endorsed by the company through analysis of customer feedback data.
Undergraduate Teaching Assistant
- Teaching Assistant for Discrete Mathematics course, offered in Spring 2021 semester by Professor Subhash Bhalla and Probability & Statistics course, offered in Monsoon 2020 semester by Professor Mahavir Jhawar.
- Responsibilities included assisting in conducting online classes, grading student submissions, conducting lab hours, and holding office hours to clarify student doubts.
Web Developer
- Redesigned and remodeled the platform’s website on WIX, increasing the user traffic by up to 20 %.
- Optimized the website’s load time and increased its SEO performance by 30 % through media optimization and extensive keyword tagging.
Research
Privacy-Preserving Machine Learning
- Mentors: Prof. Mahavir Jhawar & Prof. Debayan Gupta
- Researched and explored various methods and techniques for securely and efficiently training neural networks in multi-party settings (MPC).
- Thoroughly analyzed and implemented( in C++) the SecureNN paper for my Capstone Project
- Presentation | Report | Code
Research Intern
- Mentor: Prof. Mahavir Jhawar
- Successfully studied and implemented (in C++) different PPML (Privacy Preserving Machine Learning) protocols such as SecureML and BLAZE highlighting their merits and demerits
- Designed a faster protocol tailored to meet business requirements by developing a novel algorithm to securely evaluate non-linear functions using arithmetic shares.
- Details | Code
Secure ML & Applied Cryptography
- Mentors: Prof. Debayan Gupta for Secure ML & Prof. Mahavir Jhawar for Applied Cryptography
- Secure ML: Studied and demonstrated the impact of Data Poisoning attacks on the performance and reliability of machine learning (ML) models. Presentation | Code
- Applied Cryptography: Evaluated and illustrated the security vulnerabilities in email clients that support the two primary forms of end-to-end email encryption (OpenPGP and S/MIME) and suggested countermeasures against them. Report | Code
Projects
.js-id-ml
News Headline Generation
Explored various abstractive text summarization techniques and finetuned Google’s Pegasus model with 568 million parameters using PyTorch to generate concise headlines from summaries of local news articles. Presentation | Trained Model | Code
aCERT - Certificate Verification
Designed a blockchain-based decentralized applicatio (DApp) for pulbically publishing and verifying academic credentials Report | Presentaion | Code
India’s CSR Data Synchronization
Analyzed India’s CSR Data of the last five years using Pandas as part of the Hack & Learn Summit 2021, Government Outcome Labs, University of Oxford and presented our findings at the Social Outcome Conference 2021 (SOC21). Slides | Presentation
Synthesizing DFAs
Examined the backpropagation algorithm used to calculate gradients in recurrent neural networks (RNN) and implemented an RNN-based model in Python to automate the process of synthesizing DFAs. Presentation | Code
