If I had to capture my journey in a few words, I'd say I'm a curious guy. I'm IIT Gandhinagar (B.Tech) alumnus majoring in Computer Science and Engineering, currently pursing Master's in AI from UCLA, California. Machine Learning isn't just a discipline for me; it's a canvas where I paint with algorithms, data, and imagination. Over the years, my curiosity has led me to dive into extensive research, resulting in several published papers that mirror both my passion for ML.
Research interest: Adversarial robustness and Explainable AI (XAI) systems. During my undergrad, I was fortunate to work on multiple sub-domains of machine learning: Computer Vision, NLP & RL. Some of the problems that I have explored includes study of adversarial robustness, self-supervised learning, program synthesis, entity extraction, HDR imaging & temporal coherency in videos. Please find my papers here. However, two areas that constantly beckon me are adversarial robustness and Interpretable AI systems. There's a unique allure in ensuring the systems we design can stand their ground and, simultaneously, be transparent in their operations.
Professional Experience: I have 3 years of experience in building scalable ML systems. As of today, I am pursing my Master's in AI from UCLA. Pastly, I used to work as Machine Learning Engineer with the expertise in NLP, CV and backend development. My work with ARTPark was dedicated towards transforming the landscape of healthcare in India, where I actively contribute to innovations and solutions aimed at improving health outcomes across the nation. We are building LLM based solutions to streamline the support workflow between health workers and medical officers for faster diagnosis and treatment in rural sector of India. I have also worked at OrbitShift.ai as ML Engineer focusing mainly on LLMs & NLP. At Enphase I used to work on time-series forecasting & optimization related problems in the domain of renewable source of energy.
SZone is a B2B Augmented Reality based platform for salon and beauty franchises.
Faculty Advisor - Dr. Nipun Batra (ML) and Dr. Mayank Singh (NLP)
I throughly enjoyed teaching students about machine learning algorithms. Most of work involved designing the course syllabus, delivering practical sessions on Deep learning, developing openâsource code for student assignments and reference notebooks, weekly project discussion sessions. I mentored a total of 8 student projects in both the courses. Topics ranging from Reinforcement learning, AI Safety, Program Synthesis etc.
Faculty Advisor - Dr. Abir Das
A lot of research has been done in the field of image inpainting, however most of the research work requires drawing the object that needs to be inpainted. My task was to inpaint the object using a corresponding natural language expression. The proposed method is a two-stage architecture where segmentation mask is generated using Text-Image Architecture and inpainting is done using Wasserstein GAN.
Majorly, In South Asian countries there is a skill gap between a job seeker (15-24 yr old) and an industrial employee. And a large amount of money is spent by the government and private sector for vocational training. My job was to find major factors that leads to this skill gap and hence reduce the total amount of money spend on vocational and industrial training.
I worked on Faster RCNN model object detection model for the detection of total number of medicine strip present in a picture.
Faculty Advisor - Dr. Mayank Singh
In the recent past, we witness massive progress on the development of code generation systems for domain-specific languages (DSLs) employing sequence-to-sequence deep learning techniques. In this project we specifically experiment with AlgoLisp DSL-based generative models and showcase their extreme dataset bias through the different classes of adversarial examples. We also present a simple transformer based encoder-decoder model that outperforms the all of Algo-Lisp DSL-based baselines. However, consistent with the previous baselines, the proposed model achieves poor performance under adversarial settings.
Faculty Advisor - Dr. Udit Bhatiya
Currently, people are using deterministic mathematical models to predict rainfall and for the other related task. Since machine learning have proven to given state-of-the-art results for many different task. My task is to tackle the predictability of monsoon through machine learning algorithms.
Grant of 1500 USD for overall academic excellence.
Grant of 2500 USD as preâseed funding for startup.
Won best research proposal award at Eastern European Machine Learning Summer School(EEML-2020).
Waived registration fees and selection process for EEMLâ2021.
Implemented Satellite image segmentation model and came 3rd among all 23 participating IITs.
Build Custom Chatbot for the company website. The chatbot give the details about the company's services.
In this post we are going to take a step towards the explainability of BERT, explaining what BERT sees by analyzing the attention weights proposed in this recent paper âWhat does BERT look at? An Analysis of BERTâs Attentionâ (Clark et al., 2019).