cv

Summer Intern

Adobe Research

Working on enterprise-branded asset generation using precise trademark colors with Diffusion Transformers. Developed plug-and-play mapper for hex code to embedding translation enabling precise colorisation through prompts.

• Improved generated color accuracy by 64% over baseline methods
• Created self-curated 1200 prompt benchmark for evaluation
• Documented and submitted patent application for mapper architecture

Research Intern

Arizona State University - CORAL Lab

Working on SQL generation for queries requiring multiple tables without using LLMs for table selection. Developing end-to-end SQL generation pipeline leveraging inter-table join relationships.

• Achieved 2% accuracy improvements over LLM-based pruning baselines
• Evaluated on 3 datasets: Bird, Spider, and MMQA
• Developed intelligent table selection and pruning without LLMs

Software Intern

DevRev Cloud India Private Limited

Developed customizable NPS survey snap-in for developer organizations with automated survey distribution system. Built real-time survey update mechanisms and interval-based email dispatch system.

• Deployed system serving more than 1000 users
• Improved feedback collection efficiency through automation
• Utilized React, ReactEmail, hooks, and cron jobs for seamless user experience

Bachelor of Technology

Indian Institute of Technology Guwahati

Data Science and Artificial Intelligence

• Data Structures and Algorithms
• Database Management Systems
• Machine Learning
• Deep Learning
• Computer Vision
• Natural Language Processing
• Big Data Analytics
• Linear Algebra
• Statistical Foundations of Data Science

Inter IIT Tech Meet 13.0 - Rank 4

Inter IIT Tech Meet

Secured Rank 4 among all 23 IITs in ML/AI Problem Statement by Pathway

Encode IITG - First Position

IIT Guwahati

Team secured first position out of 2342 participants at ML hackathon across India

Amazon ML Summer School

Amazon

Selected to attend ML lectures from more than 20,000 students in India

Secretary

IITG.ai

Overall leadership of the AI Club of IIT Guwahati and its activities, promoting AI culture on campus through recruitment drives, hackathons, intensive AI/ML workshops and research paper discussions.

Agentic RAG Chatbot for LegalQA

Solved the problem of unreliable legal RAG systems that hallucinate and retrieve inaccurate information in real-time. Implemented parallel DAG-based task planning and scheduling by modifying LLMCompiler for RAG.

• Achieved 86% accuracy on CUAD dataset (contract law)
• Achieved 84% accuracy on AILA dataset (case law)
• Incorporated Pathway's vector store for real-time data processing

Adobe Behaviour Simulation Challenge

Leveraged classic ML Models, Transformers, and LLMs for behaviour simulation and content simulation on tweets. Trained DistilBERT classifier for tweet bucketing followed by separate bucket regressors.

• Achieved bucket classification accuracy of 74.3%
• BLEU-1 score of 0.13 with inference times 0.9s and 5.4s
• Generated media descriptions using BLIP-2 and finetuned Llama-2

DevRev's AI Agent 007

Engineered conversational LLM chatbot with dynamic tool selection capabilities for intelligent query resolution. Implemented pipeline integrating Retrieval-ICL, Chain-of-Thought prompting, and RAG.

• Sub-10-second query latency
• Cost efficiency under $0.02 per query
• Dynamic tool selection and sequencing capabilities