Indraprastha Institute of Information Technology, Delhi

B.Tech Computer Science and Artificial Intelligence(CSAI)

Max Healthcare & IIIT Delhi

Deep Learning Researcher

• • Continued the work from CAI.
• • Constructing and fine-tuning deep learning models for 3D medical imaging, focusing on self-supervised learning to optimise data representation. Developed a RAC-based model utilising our custom Contrastive KL Loss, allowing us to convert the high-dimensional MRI data to lower dimensions with minimal information loss.

Infosys CAI, IIIT Delhi

Machine Learning Engineer Intern

• • Worked on developing DL models for predicting segmentation masks and bounding boxes for breast cancer in patient’s DICOM MRI scans with sub-type classification. Utilised the Duke MRI Dataset along with a high quality custom dataset curated by Max Healthcare for the project.

Atlassian

Data Science Intern

• • Worked in Product Analytics Team to develop an opportunity sizing framework utilising Large Language Models(LLMs) to pinpoint initiative relevance and provide statistical context for product changes addressing business challenges revolving Jira Service Management product.
• • Wrote over 5000 lines of Python, PySpark and SQL code while utilising Databricks.

Extramarks

Data Science Intern

• • Engineered and deployed a multi-modal transformer based model for real-time assessment of student engagement during online classes utilising the video stream from students’ webcams and a comprehensive pipeline based on pyVHR, EmoFAN, and OpenVino.
• • Tech stack: PyTorch, OpenCV, TypeScript, NextJs, Firestore, GCP, and Docker.

UniRely

Full-Stack Developer

• • Engineered the frontend and backend architecture for UniRely’s website. Developed user interactions and unit-test code for robustness, including edge cases, usability, and general reliability.
• • Wrote over 8000 lines of code. Tech stack: React, NodeJS, ExpressJS, JavaScript, HTML, CSS and MySQL.

Codeforces: Maximum Rating 1707 – (Expert).

CodeChef: Maximum Rating – 2106 (5*) — India Rank under 350 (2,12,000 participants).

Guardian on Leetcode. — Among 0.3% Global Users on Leetcode.

Secured a position in the Top 50 out of 48k participants in HackOn With Amazon - Season 3.

Secured Global Rank 94 in Reply Code Challenge 2023. Qualified for Google Code Jam 2022 Round 1 and Codechef SnackDown’21 Round 1A.

Solved over 1500+ Algorithmic Problems on CodeForces, Atcoder, LeetCode, CSES and CodeChef.

Secured AIR 3681(99.68th percentile) and AIR 6347 in JEE Mains and JEE Advanced respectively.

Selected amongst a few undergraduate students across India to attend Research Week with Google 2024 in Bangalore.

Mental Health Co-Pilot(BTech Thesis) | LCS2, IIIT Delhi

• • Working on developing a conversational assistant for individuals seeking support on Online Mental Health Platforms (OMHPs). The project also involves a critical analysis of the role of conversational AI in peer-to-peer interactions within these platforms.
• • Analysed and curated a dataset with support seeker posts from seventeen mental health subreddits with appropriate annotations. Identified the critical components of an ideal support seeker post.
• • Developed a RAG and LLama2-based language model utilising our custom taxonomy for generating suggestive questions to help support seekers draft a help-seeking post. Critically analysed the model focusing on deterministic and explainable AI(XAI), and for holistic improvements using metric-triangulation strategies.

Breast Cancer Detection using DICOM MRI Scans | IIIT Delhi & Max Healthcare

• • Working on developing DL models for predicting segmentation masks and bounding boxes for breast cancer in patient’s DICOM MRI scans with sub-type classification. Utilised the Duke MRI Dataset along with a high quality custom dataset curated by Max Healthcare for the project.
• • Constructed and fine-tuned deep learning models for 3D medical imaging, focusing on self-supervised learning to optimise data representation. Developed a RAC-based model utilising our custom Contrastive KL Loss, allowing us to convert the high-dimensional MRI data to lower dimensions with minimal information loss.

Using LLMs in Software Requirements Specifications | IIIT Delhi

• • Conducted a study on fine-tuning large language models (LLMs) and prompt engineering for software development.
• • Analyzed open and closed-source foundational LLMs along with fine-tuned LLMs for domain-specific usecases to generate Software Requirements Specifications(SRS) and Software Design Documents(SDD). Worked with domain experts to evaluate the process.
• • Formulated a systematic approach to generate SRS and SDD documents using prompt engineering which allows us to develop an ideal context-prompt pair for any software project.

Real-time Student Engagement Prediction in online classes | HMI Lab, IIIT Delhi

• • Curated a custom dataset for training transformer based multi-modal solutions for engagement prediction in online classes. Utilised concepts from Affective Computing such as the VAD (Valence-Arousal-Dominance) model to identify relevant modalities, formulate training losses and methodology.
• • Developed an ensembled SwinT based model performing better than the current SOTA with 5 classification heads to improve accuracy for classes with lower frequencies due to long-tail nature of the dataset.
• • Further worked on reducing the inference time latency of the model for real-time utilisation.