Hi, I'm Rahul Vishwakarma

Currently (April 2025 – Present), I am working as a System Engineer at BHERI Tech Foundation, where I am developing LLM-based agents to answer questions related to Indian heritage, referencing primary sources from the Dhara platform. The current version of the service is available at: Dhara Prashna Page.

Previously (September 2024 – March 2025), I worked as a Junior Research Fellow (Project Associate) at IIT Hyderabad under the guidance of Dr. Mohan Raghavan. Our project focused on building a digital library to preserve and share ancient Indian knowledge by scraping diverse sources and creating a public-facing interface. The web portal providing access to our three main services (Dictionary, Verse Finder, Chunk Server) is available at: Dhara Page.

This was my MSc thesis, supervised by Dr. Subhankar Mishra. In this research, we fine-tuned LLMs (ByT5) and combined them with the interactive prover Lean to automate mathematical proof generation. Some of our contributions are - introduction of a dynamic sampling method, augmentation of the training dataset, and development of a tokenizer specific to Lean 4. Additionally, we developed a website for Neural Theorem Proving to make our method easily accessible through both web and API usage.

In this project, we focused on improving the accuracy of location prediction in an indoor environment using WiFi RSSI. We trained machine learning models to capture intricate signal strength patterns unique to indoor spaces. We introduced a method called IndoorGNN, which utilizes Graph Neural Networks to achieve superior indoor localization performance, surpassing conventional algorithms such as kNN, SVM, and MLP. This project resulted in the publication of IndoorGNN.

In this project, we used machine learning for change detection (for roads and buildings) in satellite images over time. We trained ML models to predict masks for roads and buildings, then compared them across different timelines to identify changes. We also launched a website where users can upload two images of a region to detect changes using our trained model.

During my internship at IISc Bangalore with Prof. Siddhartha Gadgil, I developed skills in theorem proving using the Lean Interactive Theorem Prover. Additionally, we explored the use of machine learning models to predict proof steps for theorems.

Developed a file access pattern-based recommendation system for NISER Archive to suggest study materials.

Developed a web app for sharing the live location of the NISER buses with its members.