Back
Projects
This is my thesis, Evaluating efficiency of LLMs in the field of Computational Social Science, developing a scalable Chain-of-Thought annotation pipeline to classify 3.5M+ high-risk posts (~80% manual acceptance accuracy) and developing agentic workflows with Temporal Knowledge Graphs for bias-aware, interpretable socio-political event prediction.
An experimental course project to understand use-cases of Graph Neural Nets, developed predictive models using GNNs to determine aqueous solubility of chemical compounds, identifying potential for advancing molecular property predictions in drug discovery and environmental chemistry.
A paper and a project. A complete end-to-end NLP pipeline using a Custom Neural Network and word embeddings, achieving 93% and 98% accuracy in detecting fake news in Hindi and English, with a custom dataset of 10,000 articles and a functional user interface.
My first try in fine-tuning LLMs on Apple Silicon. Optimized the Llama-3 8B Instruct model by fine-tuning on SEC filings data to build an RAG pipeline using Langchain, delivering actionable insights from 10-K reports.
A deep dive into an exciting research project as a part-time researcher at IIT Madras, built a simulation of the South Indian railway network. Scraped track lengths, station data, and train timetables from railway websites, cleaned and structured the data while standardizing key terminologies. Developed a database pipeline linking station routes and modeled real-world train operations using SimPy. The simulation incorporated parameters like cruise speed and signaling protocols, running actual timetables on the network and generating detailed activity logs for further analysis.
My final undergrad project where I created an emotion classification solution from EEG signals using Symbolic Regression, utilizing features from Discrete Wavelet Transform-derived frequency bands, achieving 84.39% accuracy.