Jay Kulkarni
  • Home
  • About
  • CV
  • Projects
  • Blog

On this page

  • Abstract
  • Authors
  • Poster Presentation
  • Replication

Information Extraction From Fiscal Documents Using LLMs

Public Finance
AI
India
XKDR Working Paper
Authors

Vikram Aggarwal

Jay Kulkarni

Aakriti Narang

Aditi Mascarenhas

Siddarth Raman

Ajay Shah

Susan Thomas

Published

November 13, 2025

Abstract

Large Language Models (LLMs) have demonstrated remarkable capabilities in text comprehension, but their ability to process complex, hierarchical tabular data remains underexplored. We present a novel approach to extracting structured data from multi-page government fiscal documents using LLM-based techniques. Applied to large annual fiscal documents from the State of Karnataka in India, our method achieves high accuracy through a multi-stage pipeline that leverages domain knowledge, sequential context, and algorithmic validation. Traditional OCR methods work poorly with errors that are hard to detect. The inherent structure of fiscal tables, with totals at each level of the hierarchy, allows for robust internal validation of the extracted data. We use these hierarchical relationships to create multi-level validation checks. We demonstrate that LLMs can read tables and also process document-specific structural hierarchies, offering a scalable process for converting PDF-based fiscal disclosures into research-ready databases. Our implementation shows promise for broader applications across developing country contexts.

Clickable Button Access paper on ARXIV Access paper on REPEC

Authors

Vikram Aggarwal, Jay Kulkarni, Aakriti Narang, Aditi Mascarenhas, Siddarth Raman, Ajay Shah, and Susan Thomas

Poster Presentation

Presented at ACM ICAIF 2025 in November 2025.

Unable to display PDF file. Download instead.

Replication

A replication package is maintained on the Github repository ‘LLM_table2db’ with input documents, source code, and prompts.

Clickable Button Github: LLM_table2db
Copyright 2025, Jay Kulkarni