CV

Bachelor's in Computer Science and Engineering

Kalinga Institute of Industrial Technology

GPA: 3.60

Master's in Computer Science with Specialization in Machine Learning

George Mason University

GPA: 3.93

Research Assistant

George Mason University

• Working with Prof. Ziyu Yao in the domain of Information Extraction, more briefly explored generalization in Event Extraction tasks through instruction-tuned LLMs aided by diverse set of annotation guidelines.
• Submitted research paper Instruction-Tuning LLMs for Event Extraction with Annotation Guidelines to ACL 2025.
• Improved F1-score by 10% in EE tasks by synthesizing structured guidelines for 500+ events and 4000+ arguments, and fine-tuning LLaMA-3.1 8B using LoRA and regularization techniques.
• Designed a custom evaluation script to ensure robust benchmarking for Event Argument Extraction (EAE), Event Detection (ED), and End-to-End (E2E) EE tasks.

Summer Research Intern

George Mason University

• Working with Prof. Ziyu Yao in the domain of Information Extraction, more briefly explored generalization in Event Extraction tasks through instruction-tuned LLMs aided by diverse set of annotation guidelines.
• Reduced hallucinations and enhanced evaluation consistency by developing a preprocessing pipeline that converts unstructured text into structured Python code prompts.
• Enhanced multi-domain generalization by constructing an ontology for event and argument identification, applicable to 3 EE tasks across 8 domains.
• Enhanced model interpretability and performance by generating synthetic annotation guidelines for 500+ events and 4000+ arguments using GPT-4, enabling more consistent event extraction.

Data Scientist

Highradius Technologies Private Limited

Implemented and deployed machine learning models for Fortune 500 companies, enhancing operational efficiency and predictive accuracy.

• Increased dollar recovery rate by 12% implementing imbalanced classification ML model (DVP) for fraud detection with 85% average accuracy for 10+ fortune 500 clients.
• Increased sales acceleration and reduced business disruptions by 20% through the optimization of credit decisioning for order release using predictive models.
• Reduced Days Sales Outstanding (DSO) by 30% by building a regression model for Payment Date Prediction (PDP).
• Accelerated DVP and PDP deployment by 48% and 57%, automating the model building and deployment pipelines.
• Productionized ML models for 10+ Fortune 500 clients like Nestle, Kraft Heinz, Clorox, Uber, Kellogg’s, etc.
• Enhanced stakeholder decision-making by visualizing model insights and business impact using Tableau.
• Developed a digital assistant, Freeda, used by Fortune 500 companies to understand intent behind user queries and provide appropriate responses for multiple Highradius cloud products.
• Mentored over 15 interns in data science and business concepts during the 2021 and 2022 internship programs.

Research Intern

Indian Institute of Technology, Kharagpur

Developed BCH error correction codes and tested OpenSSL authentication protocols.

• Implemented BCH error correction code and compared existing test suites in OpenSSL with the proposed authentication protocol.
• Published a research paper titled 'An OpenSSL Extension for PUF-based Authentication' in IEEE DSP 2018.

Project: Medi-Buddy - Multi-Agent Framework for Medical Insights

Developed a multi-agent framework using LLaMA-3.3 70B to extract and summarize evidence-based medical insights from PubMed and Wikipedia.

• Integrated multiple data sources including PubMed, Wikipedia, and DuckDuckGo via APIs.
• Leveraged Groq’s inference API and Phidata orchestration to create a streamlined agent workflow.
• Focused on delivering accurate, explainable, and domain-specific responses in the medical domain using cutting-edge LLMs.

Research Project: Event Extraction

Enhancing Event Extraction using instruction-tuned LLMs to achieve superior performance on unseen domains.

• Improved F1-score by 10% in EE tasks by synthesizing structured guidelines for 500+ events and 4000+ arguments, and fine-tuning LLaMA-3.1 8B using LoRA and regularization techniques.
• Reduced hallucinations and enhanced evaluation consistency by developing a preprocessing pipeline that converts unstructured text into structured Python code prompts.
• Enhanced multi-domain generalization by constructing an ontology for event and argument identification, applicable to 3 EE tasks across 8 domains.
• Designed a custom evaluation script to ensure robust benchmarking for Event Argument Extraction (EAE), Event Detection (ED), and End-to-End (E2E) EE tasks.

Project: Multi-Cloud AI-Driven DevOps

Designed a scalable, AI-powered e-commerce platform using infrastructure and services across AWS, GCP, and Azure.

• Provisioned infrastructure using Terraform with AWS EC2, EKS, and DynamoDB.
• Containerized and deployed frontend/backend apps on AWS EKS using Kubernetes.
• Built CI/CD pipeline with GitHub and AWS CodePipeline for seamless deployments.
• Integrated Amazon Bedrock for product recommendations and OpenAI GPT-4o for customer support.
• Connected AWS Lambda to Google BigQuery and used Azure AI for sentiment analysis.
• Enabled smooth cloud-to-cloud data exchange ensuring system resilience.

Project: Gentopia-Mason - Enhancing LLM Agents with Currency Conversion & PDF Reader

Extended Gentopia to support real-time currency conversion and document extraction with LLM-powered agents.

• Built Currency Conversion Agent using Fixer.io API for real-time FX data.
• Developed PDF Reader Agent using PyPDF2 for extracting structured information from documents.
• Enabled financial, document, and general-purpose NLP capabilities using modular agents.

Project: Counterhate Arguments - Enhancing Online Discourse Against Hate Speech

Reproduced and extended EMNLP study on counterhate arguments to tackle online hate speech effectively.

• Implemented full NLP pipeline for data preprocessing, model training, and evaluation.
• Trained and fine-tuned models like RoBERTa and Longformer for classification tasks.

Project: YouTube Global Statistics Analytics

Developed an interactive web dashboard to visualize global YouTube trends across countries and categories.

• Built six interactive visualizations using Plotly including bar charts, line graphs, heatmaps, and choropleth maps.
• Implemented a global slider filter for dynamic data analysis by view thresholds.

Project: Multilingual Sentence Classification for Media Framing and Bias Detection

Fine-tuned BERT for multilingual sentence classification to detect media bias with 15-class taxonomy.

• Achieved 70% classification accuracy using 450+ annotated sentences.
• Applied data augmentation techniques like translation and paraphrasing to boost model robustness.

Project: Dynamic Bus Routing System

Developed an algorithmic solution to alleviate congestion on specific routes through dynamic allocation of buses.

• Designed a system that dynamically allocates buses based on capacity (via ticketing system) and demand within each route.
• Implemented YOLO object detection to calculate demand at each bus stop and inside buses.

Project: Dynamic Traffic Management System

Created an algorithmic approach to optimize traffic light timing in urban environments.

• Used traffic flow analysis to reduce congestion at key intersections.

Best Performer of the Team

Highradius Technologies

Awarded for exceptional contributions across three quarters.

Winner, Smart India Hackathon: National Level Hackathon by Govt. of INDIA

Government of INDIA

Recognized for developing an algorithmic solution to alleviate congestion on throughout all routes within a city through dynamic allocation of buses.