About
My Introduction
I am deeply passionate about creating intelligent machines that enhance human capability and accessibility. With a strong background in artificial intelligence, machine learning, and data mining, I enjoy tackling complex challenges and transforming them into practical, human-centric solutions. My experience spans natural language processing, retrieval-augmented generation, and vector databases, where I focus on building systems that make a real-world impact. Through three internship experiences in the computer science domain and rigorous coursework and project work, I have honed my technical skills and problem-solving abilities. Driven by a commitment to continuous learning and hands-on development, I strive to design intelligent systems that empower individuals and promote inclusivity.
Skills
My ExpertiseExperience
Professional and Academic ExperienceTeaching Assistant for CS6220 Data Mining
Khoury College of Computer Sciences, Northeastern UniversityMS in Artificial Intelligence
Foundations of AI, LLMs, Algorithms, Machine Learning | Northeastern University |Boston, USABTech in Computer Science (Spz AI&ML)
Software Engineering, NLP, AI, ML, Data Mining, OOP | UPES | Dehradun, IndiaAI Research Engineer
SignoFiSoftware Engineer - R&D
KEKSoftware Engineer Intern - Knowledge Graphs
Jio-RelianceProjects
Some of my projects
ClinAI: MCP based Agentic AI for healthcare Clinical assistant
ClinAI is an agentic AI system that automates the extraction, organization, and retrieval of clinical data from doctor-patient conversations using large language models. Built on the Model Context Protocol (MCP) framework, ClinAI enables seamless coordination between multiple AI agents and tools. It uses Google Gemini for post-session speaker role labeling and intelligent summarization of transcriptions. The workflow involves real-time audio capture, structured note generation, editable record validation, and persistent storage via MongoDB. ClinAI demonstrates the potential of Agentic AI to reduce administrative overhead, ensure data traceability, and deliver scalable, privacy-conscious healthcare automation.
GitHub
Machine Unlearning: Erasing concepts from LLMs
In this project, we aim to enable large language models to selectively forget specific information without requiring full retraining. We compare three different approaches — LoRA-based fine-tuning, guardrailing using LLaMA-Guard, and Sparse Autoencoder (SAE)-based unlearning — to assess their effectiveness in targeted forgetting. The workflow includes curated dataset preparation, parameter-efficient training, evaluation through perplexity and retention metrics, and runtime safety checks. This work highlights practical, scalable techniques for efficient and legally compliant machine unlearning.
GitHub
BlindSight: AI-Powered Voice-Driven File Navigation for the Visually Impaired
BlindSight is our innovative approach to making computer systems more accessible and intuitive for individuals with complete or partial visual impairment. Using just a single key, users can navigate the operating system and perform essential tasks like text-file editing and OS manipulation (e.g., changing directories, listing files, creating folders, etc.). Our mission is to foster inclusivity in the tech space and empower individuals with disabilities by providing them with tools for equal representation in the digital world.
GitHub
AI-Powered Grading Assistant
In this project, we develop an AI-powered grading assistant designed to automate score prediction and feedback generation for short answers. The system fine-tunes RoBERTa for numerical score prediction and BART for detailed feedback generation using parameter-efficient LoRA methods. The training process involves curated dataset preparation, hyperparameter tuning, and evaluation using metrics such as RMSE, Pearson correlation, and ROUGE scores. This work demonstrates how fine-tuned transformer models can streamline grading tasks by delivering accurate, explainable assessments at scale.
GitHub
Anakin Breaking Bad
Developed a fun game based on Star Wars. Used Reinforcement learning techniques like value iteration, Natural language processing, Text similarity and Pygame. Do check it out and give suggestions!
GitHub
DCGAN vs VAE - A comparison for Image Generation
This project explores Deep Convolutional Generative Adversarial Networks (DC- GANs) and Variational Autoencoders (VAEs) by developing and comparing them with the CIFAR-10 dataset. Utilizing the PyTorch library, these models are built from scratch to generate and reconstruct images. The study evaluates their perfor- mance using the Inception score (IS) and the Fréchet Inception Distance (FID).
View Paper
Memory Companion
MemoryCompanion is a tool for people suffering with Mild Cognitive Impairment (MCI), a condition in which users face memory or thinking related problems but not to the point that it affects their daily activities. These people tend to miss social events, forget appointments and have issues with remembering names or conversations with other people. MemoryCompanion helps these people keep track of conversations, names, tasks to do and much more!
GitHub
Surface Crack Detection
Built a surface crack detection web-app with a lightweight custom CNN to help civil inspectors inspect concrete surface quality in earthquake prone areas.
GitHub
Agent Navigation using reinforcement algorithms (comparative study)
Reinforcement learning has been widely used as a learning mechanism for an artificial life system. I built a project that compares path planning algorithms like Q- learning, SARSA, DQL (Path planning via deep reinforcement learning is an end-to-end method) in terms of optimality and time taken. I also visualize the path planned by each one of them using simple graphics.
GitHubWant to collaborate?
Feel free to reach out if you're interested in collaborating on projects or research, particularly in the fields of AI, Data Science, ML, GenAI, or related areas!
Contact
Blog
Explore My Medium ArticlesStay updated with my thoughts, insights, and research in AI, Data Science, ML, and more on Medium!
Visit My Blog