Projects
A selection of things I’ve worked on. Code is on GitHub.
Multi-Agent Framework for Generating Executable Research Papers
A domain-agnostic autonomous multi-agent system built on the OpenAI API that orchestrates planning, code generation, and execution to turn research questions into fully executable scientific papers. Includes automated environment construction, sandboxed execution, logging, and a reproducibility fault-injection pipeline. Generated and evaluated 30 synthetic papers across domains, outperforming a single-LLM baseline (24.1 vs 14.1 avg. score).
Marathi Paraphrase Generation
Analysed 4 LLMs for generating Marathi paraphrases. Curated a dataset of Marathi phrases and their paraphrases, and evaluated model performance using BLEU, METEOR, cosine similarity, and Levenshtein distance.
Spotify Playlist Recommendation
Built as part of the Microsoft Engage Mentorship program — a KNN-based machine learning model that predicts the most suitable playlist for a song, paired with a Django frontend integrated with the model.