Experiences
Overview of My Professional Journey.
Developed an automated work planning pipeline using JavaScript, boosting operational efficiency by over 70%.
Performed financial maturity assessments using Qualtrics surveys and Tableau to benchmark client performance and recommend strategic improvements.
Conducted benchmarking analysis for major clients by extracting and processing data from proprietary sources.
Built a JavaScript add-in to enhance website usability by enabling persistent default option settings.
Built and fine-tuned custom GPTs on open-source datasets to support internal research and domain-specific tasks.
Created an Excel add-in "GoToColumn" using TypeScript to let clients access and manipulate columns easily.
Utilizes the Office JavaScript API for interacting with strongly-typed Excel objects including worksheets.
Developed a system to parse and consolidate bank statement PDFs into Excel using Python and Regex.
Increased productivity by reducing manual lookups through replacing Power Query Editor with Python code.
Projects
Here's what I have built throughout my journey.
Crafted a robust real-time collaborative platform with integrated chat, coding, and sketching features.
Implemented auto-language detection and in-browser code execution using CodeMirror component.
Enabled WebSocket communication for instant messaging and real-time notifications on join/leave events and activities, ensuring security and integrity.
User-friendly application for API testing and interaction, offering streamlined workflows for newbie developers.
Implemented request history and repeat requests features, making API testing more intuitive.
Provided code snippet generation for Axios and Fetch, simplifying coding with ready-to-use templates.
Enhanced Excel user experience with advanced column navigation and management, built using Typescript.
Significantly reduced time spent accessing columns across spreadsheets by over 50% while boosting accuracy and productivity.
Implemented features include hide/unhide, sheet locking, auto-refresh, and column profiling via Excel JavaScript API.
Developed a Streamlit application for file upload, text extraction, and conversational querying using an AI chatbot.
Text extraction using PyPDF2 and python-docx, text chunking for OpenAI GPT-3.5 model input limitations
Designed a user-friendly interface for seamless interaction and improved user experience
Team project for showcasing university-wide events and handling room booking.
Developed user and admin signup/login page and linked it with the backend to store credentials in encrypted form.
Developed infinite scroll sponsorship section for enhanced user experience.
Research And Publications
Overview of My Research Journey.
Promoting Fairness in LLMs: Detection and Mitigation of Gender Bias
Designed a bias evaluation and mitigation pipeline for Large Language Models (LLMs) using prompt engineering and LoRA-based fine-tuning on a custom fairness-focused dataset.
Introduced specialized metrics—Disparity Index (DI), Idea Consistency Score (ICS), and Thematic Consistency Score (TCS)—to measure bias across gender, race, and other sensitive dimensions in English and Hindi.
Achieved significant reduction in bias: 37% (gender), 27% (race), and 30% (age) using LoRA fine-tuning on the LLaMA 3B model.
Used GPT-4o to generate 170 curated prompt-response pairs across themes like social justice, cultural representation, and accessibility; publicly released via GitHub.
Evaluated models through implicit association testing (IAT), zero-shot classification, and real-world scenario prompts, highlighting effectiveness across languages and demographic factors.
Semantic Textual Similarity with Supervised and Unsupervised Learning
Developed a hybrid model combining Supervised and Unsupervised Learning approaches for Semantic Textual Similarity (STS) tasks.
Utilized SVR, LightGBM, and XGBoost models, alongside feedforward neural networks, to achieve enhanced text similarity predictions.
Achieved industry-standard performance with a Pearson correlation of 0.84, Spearman correlation of 0.82, and MSE of 0.73 by leveraging novel ensemble techniques integrating SVR and Neural Networks.
Conducted experiments on benchmark datasets (SemEval 2012 Task 6), achieving robust cross-validation results across 5 folds for 24 candidates in 120 fits.
Pinpointed key factors impacting student success, using data visualization libraries and analytics.
Analysed patterns linking academics, activities, internships, etc, for better career choices.
Achieved an impressive accuracy of 0.81 by implementing advanced ensembling techniques, contributing to enhanced predictive modelling capabilities for student success initiatives.
Tech Stack
Languages, libraries and frameworks I know and use.