PersonaRAG is a production-grade Retrieval-Augmented Generation (RAG) system that acts as a conversational AI twin of my professional profile. It answers questions about my experience, projects, research, and education in my own voice, grounded strictly in my resume and personal website. Unlike standard RAG pipelines, PersonaRAG is optimized for multi-turn conversations and resume-style content, where users expect complete sections (e.g., “Work Experience”) and coherent follow-ups rather than isolated text snippets.

Implementation of RAG method to perform question answering on PDF documents. HuggingFace embeddings in addition to the Groq API were used for efficient and accurate information retrieval. Session management was done to ensure embeddings are computed once per document upload to optimize performance by avoiding redundant computations.

Development of Web App that uses open source LLMs to answer user queries based on its own knowledge as well as any reference provided in the form of external webpages.

"Forecasting the Future: Utilizing LSTM for Next-Word Prediction in Natural Language Processing."

Development of a Machine Learning Model that takes user's activity data as input and predicts if the user is going to churn

Use image processing and deep learning to alert users at the frontend that their photos might be rejected in the backend as their photo do not meet the requirements such as lighting, eyes properly visible, no spoof and properly clothed.

Development of models that extract parameters like account number, ifsc, date, name, duration and so on and matches them with the requirements to accept or reject the account statement as a bank or income proof

Development of Machine Learning Model that predicts similarity between 2 names trained on data collected for indian names.

A KUKA robot was automated using a simple camera that enabled to locate, localize, identify the pose and feed the coordinates to robot in order to grisp it. Image processing techniques and deep learning methods were used recognize the object and PnP model was used to find the pose for a perfect grasp.