I'm Niraj Kumar. Researcher, Data Scientist and a Software Engineer with a passion for uncovering insights through data and creating intelligent software solutions that help businesses make informed decisions. I invite you to explore my work and learn more about my journey in technology.
Currently, I'm pursuing my Ph.D in Image Processing from University of Delhi. I have completed Master's in Computer Applications from Vellore Institute of Technology, Vellore Tamil Nadu, and hold a Bachelor's from the University of Lucknow. I gained hands-on experience as a Data Scientist Intern at DeepThought EduTech Ventures Pvt. Ltd., where I developed an NLP module for Virtual Project Showcase Platform and designed a dashboard for the UBS(Unicorn Behavioral Score), transforming stored employee behavioral data into a centralized, actionable overview. I also co-led a research project to enhance engineering education. I have strong proficiency in various programming languages and have developed skills in advanced areas like Deep learning and Generative AI. I excel at creating insightful data visualizations and have explored various backend web development frameworks and tools. Additionally, I have learned MLOps for the efficient deployment and management of machine learning models and am well-versed in large language models.
Python, C++, SQL, Java, R, JavaScript
Machine Learning, Deep Learning, MLOps, Generative AI,
NumPy, Pandas, Scikit-Learn, TensorFlow, PowerBI, Figma, Excel, Langchain, Docker
Node.js, Express.js, React.js, Django, PostgreSQL, VectorDB
This project focuses on building a tool that allows users to ask questions directly from PDF documents and get meaningful answers. Using RAG technology with chat history, the system retrieves relevant information from the document and generates accurate responses. It employs Gemma2-9b-It for large language model capabilities. Users can upload PDFs and interact with their content through a simple chat interface. It’s particularly useful for businesses dealing with large volumes of documents, such as legal firms or research teams, who need quick insights without manually sifting through pages.
This module is part of a virtual project showcase platform. By integrating the SlamBook Machine Learning module, which utilizes LSTM-RNN(Long Short-Term Memory Recurrent Neural Networks), it generates personalized feedback from Compliments, Questions, and Constructive Feedback. This AI-driven tool not only recognizes student achievements but also encourages growth, delivering a memorable digital souvenir. As a unique platform, it enhances student engagement and provides institutions with a competitive edge in showcasing their students' accomplishments effectively. This module is deployed on Streamlit.
This project presents a dynamic Power BI dashboard analyzing sales data for ElectroHub, a retail company offering diverse products. Using SQL Server and Excel, it features sales trends, top/bottom products, profit insights, discount analysis, and geographic sales distribution. The dashboard uses a star schema data model and interactive tools like slicers and drill-throughs to deliver actionable insights. It supports informed decision-making by optimizing promotions and identifying high-performing products and regions effectively.
Designed and implemented a highly accurate regression model to predict house prices in Chennai, utilizing extensive real estate data that included variables such as location, size, amenities, and historical price trends. The model was fine-tuned to capture the nuances of the Chennai real estate market, providing precise price predictions that aid both buyers and sellers in making informed decisions. This solution was then transformed into a user-friendly web application using Flask, ensuring easy access and interaction. The application was containerized with Docker, and deployed on Render.
Developed a Convolutional Neural Network (CNN) model to mimic human driving behavior, enabling autonomous navigation in a simulated environment. The project utilized Computer Vision techniques for real-time image processing and decision-making. Flask and Socket.io were employed to create an interactive interface for real-time communication between the simulation and the model, allowing for control and visualization of the autonomous driving process.
By focusing on optimizing demand and supply chain management, developed a high-performance sales prediction model that empowers businesses to make data-driven decisions on inventory management and resource allocation. Leveraging ensemble techniques like Random Forest and Gradient Boosting, the model achieved significant improvements in accuracy and stability, ensuring reliable sales forecasts. ancd created simple Dashboard for Analysis
Traveller is a web platform where users can explore and manage their favorite travel destinations. Built with a basic frontend and a Django backend, it uses PostgreSQL for data storage and is deployed on Render. Users can register, log in, and view their travel plans, while only admins have the ability to add or remove destination data. The platform provides personalized travel experiences and offers collaboration opportunities with travel agencies, helps users plan their trips while businesses can promote curated experiences and services.
Published in IEEE on April 18, 2024.
Signature is a powerful identity of a person for various legal acts, to authorize a transactions to avoid financial crimes. Now-a-days, we are using touch-sensitive pad for the verification of the signatures. Thus there is high risk of forgery to happen due to replicate the signature patterns. So it is essential to detection and verify the signatures while using public platforms to avoid crime of accessing the personal data. The advancements in image processing and artificial neural network helps in identifying and validating genuine signature.
This research focused on identifying and predicting forged handwritten signatures using advanced computer vision techniques. This paper proposed a signature validation process using harris corner detection and Siamese neural network. Harris corner detector helps to extract the corners and infer the features of an signature image, whereas Siamese network identifies the pattern of signature by learning with the help of similarity function by taking the probability of the signature images.