PySpark Visual Workflow Builder
A data engineering project to create a drag-and-drop workflow builder for Apache Spark pipelines. The system allows users to visually construct Spark RDD workflows using a Node-RED frontend, a Flask backend, and Apache Spark for execution. It enables interactive design and seamless execution of distributed data processing tasks, lowering the barrier to big data development.
ArttMart - E-commerce Platform
A full-stack e-commerce application built with Next.js, Redux, Node.js, Express.js, and PostgreSQL, featuring secure JWT-based authentication and a seamless user experience for browsing artwork, searching products, managing carts, placing orders, and making payments. Implemented CI/CD pipelines with GitHub Actions and Docker for automated testing and continuous deployment. The backend is containerized with docker and deployed on Oracle compute instance, enabling zero-downtime updates.
PlotPact
An applied AI product designed to support creative writing through interactive storytelling. PlotPact integrates large language models (LLMs) with constraint-based guidance, allowing users to co-create stories while adhering to specific narrative rules. Built with Next.js, Flask, and OpenAI APIs, the platform enables real-time collaboration between human creativity and AI assistance.
Blogging Web Application
A full stack project to develop a blogging application using modern web technologies and libraries like Next.js, Editor.js, Material UI, Node.js, Express.js and MongoDB. The application allows user to register with Email verification, login with JWT, create, update, read and delete blogs, and upload profile images enabled by Cloudinary.
Desktop Accounting Application
A freelance project to develop an accounting application for Regency Books and Stationary, Birtamod. A desktop application made with Node.js, PostgreSQL, Electron and React-Admin. App has features like order and return management, billing, ledger management, receipt generation and a dashboard with various charts.
Research Projects
NESA: Multi-Agent LLM Essay Evaluation
A four-agent CrewAI pipeline that turns LLM-as-judge essay scoring into a propose, justify, challenge, and adjudicate debate, improving agreement with expert human raters by 7% QWK over zero-shot baselines. It also introduces NESA, the first expert-annotated Nepali automated-essay-scoring benchmark, and benchmarks three open-source LLMs across single-prompt and agentic workflows. Published at IEEE BigData 2025.
ClimaConvo: Climate change twitter data analysis
A comprehensively annotated multi-aspect dataset pertaining to climate change discourse in twitter with analysis of the tweets through advanced topic modeling and sentimental analysis techniques. Conducted experiments with various architecture and established baselines for 6 tasks. Paper submitted to LREC-COLING 2024
