cv

Research Intern

CSE Department, IIT Bombay

Working on a project on advancing Automatic Speech Recognition (ASR) systems on low-resource accented speech under the guidance of Prof. Preethi Jyothi.

• Automatic Speech Recgnition
• Low-resource NLP

Student Researcher

APPCAIR, BITS Pilani

Worked on the extension of CRMs paper by Prof. Srinivasan and Tirthraj Dash, a form of 'explainable neural networks' by large-scale pretraining and examining their behavior.

• Symbolic AI
• Autoencoders

Student Researcher

APPCAIR, BITS Pilani

Under the guidance of Senior Prof. Ashwin Srinivasan, Dr. Lovekesh Vig, and Dr. Gautam Shroff, I worked in collaboration with TCS research to study and analyze the performance of LLM's ability to reason over arguments by introducing Argumentative Reasoning.

• Large Language Models
• LLM Agents

Summer Intern

Dino.co.in

Demonstrated initiative and managed project independently to develop a GPT-powered chatbot using LangChain to facilitate content creation and enhanced customer support

• LangChain
• prompting

Core Member

SAiDL

SAiDL is a student-led research group at BITS Pilani, Goa, working on research and applications of various fields in ML. I am a member of the group and have worked on various projects, organized events and conducted courses.

B.E. (Hons.)

BITS Pilani

Electrical and Electronic Engineering (Minor in Data Science)

• Machine Learning
• Large Language Models
• Foundations of Data Science
• Optimization
• Deep Learning

IKDD Uplink Research Internship

IKDD

One among 12 students in the country to be selected for a 3-month long research internship.

LLM Hackathon

Moveworks

Second position amongst 20 shortlisted teams that participated, having around 80 people in total.

CountCLIP

The project involves reproducing the ICCV 2023 paper 'Teaching CLIP to Count to ten', ensuring implementation accessibility, and using a specialized dataset for training.

Rank-N-Contrast for graphs

Reproduction of the NeurIPS Spotlight Rank-N-Contrast, and evaluating performance in graph regression tasks.

Albert with Perceiver layers

The project involved implementing the Albert model with Perceiver layers and comparing its performance against the standard Transformer layers. Both models were pre-trained on the same corpus and evaluated through fine-tuning for paraphrasing tasks using the MSR corpus.

Code-Mixed Sentence Generation and Language Model Fine-Tuning

The project involved examining code-mixed sentences with non-formal language for abuse detection. Pre-trained language models (BeRT and m-BeRT) were fine-tuned to categorize code-mixed sentences and assess their performance.

Zero-Shot Image Segmentation using CLIP

The project achieved text-guided image segmentation by leveraging CLIP’s text-image embeddings. The approach utilized a contrastive loss to align with ground truth segmentation maps.

Variations of Softmax

Project analyzed how various Softmax variants affect both model performance and training time, evaluating them on large-class classification tasks. This explores the trade-offs between computational complexity and model accuracy to enhance computational efficiency