At Pune Institute of Computer Technology, I built a Python pipeline for Marathi Legal NLP that converted 6,744 court-judgment PDFs into anonymized documents. I fixed broken Marathi font encodings and reduced invalid word starts by 99.5%.
I also cleaned and deduplicated the corpus, segmented documents into six section types, and added metadata extraction. I developed rule-based anonymization with leak scanning and Pytest regression tests.
On IP-Sakti (AyurX), I built a citation-grounded RAG pipeline over legal and pharmacopoeial PDFs, combining Qdrant retrieval with BM25 and cross-encoder reranking. I evaluated retrieval and engine correctness separately from end-to-end LLM answer quality.
My projects also include fine-tuning SegFormer for terrain segmentation and building a TensorFlow/Keras pneumonia detection web app. I ranked among the top 7 of 3,722 teams at the MIT AI Grand Challenge and was a finalist at the IEEE and YOLO hackathons.

