Sumantra Kumar

About Sumantra Kumar

An aspiring Data Scientist / Analyst, currently pursuing Masters in Applied Statistics. A…

Experience

1. Data Scientist

   LSEG

   Sep 2023 — Present · Bengaluru, IN

   Automated Data Extraction from Unstructured PDFs:Developed a scalable solution using PyMuPDF and regex to extract structured data points from PDFs, significantly reducing manual processing- Table of Contents (ToC) Generation via Layout Analysis:Built a pipeline to extract section headers, table titles, and figure captions from PDFs by fine-tuning a layout transformer. Used OCR on bounding boxes and created an auto-tagging script to speed up training data creation- Speech-to-Text with Speaker Attribution:Deployed OpenAI’s Whisper model to transcribe audio files and integrated speaker diarization for clearer, multi-speaker transcripts- Web Data Extraction Using Markdown + LLMs:Scraped company websites using BeautifulSoup and used quantized 7B LLMs (via llama.cpp) with prompt engineering for efficient, high-quality attribute extraction- RAG System for PDF Attribute Extraction:Built a cost-efficient Retrieval-Augmented Generation (RAG) system to extract attributes from large PDFs by providing concise, context-aware inputs to LLMs, resulting in high-precision output- Vision-Based PDF Parsing for Tables & Figures:Used vision-language models and prompt engineering to accurately extract data from embedded tables and figures within PDFs- Additional Work in Web Scraping & Model Hosting:Delivered other web scraping/extraction projects and deployed both open-source (LLaMA 3 series, Mistral 7B) and proprietary models (GPT-4o, GPT-4o mini) for varied use cases.

Education

• Delhi University

  Bachelor of Science - BS, Statistics

  2017 — 2020

• Symbiosis Statistical Institute, Pune

  Master of Science - MS, Applied Statistics

• Ryan International School , Mayur Vihar , Delhi

  10th - 12th Class , PCM

  2017