Despite comprehensive food safety regulations worldwide, violations continue to pose significant public health challenges. This paper presents an LLM-driven pipeline for analyzing legal texts to identify structural and procedural gaps in food safety enforcement. We develop an end-to-end system that leverages Large Language Models to extract structured entities from legal judgments, construct statute-and-provision-level knowledge graphs, and perform semantic clustering of cases. Applying our approach to 782 Indian food safety violation cases filed between 2022-2024, we uncover critical insights: 96% of cases were filed by individuals and organizations against state authorities, with 60% resulting in decisions favoring appellants. Through automated clustering and analysis, we identify major procedural lapses including unclear jurisdictional boundaries between enforcement agencies, insufficient evidence collection, and ambiguous penalty guidelines. Our findings reveal concrete weaknesses in current enforcement practices and demonstrate the practical value of LLMs for legal analysis at scale.

Legal documents are complex in nature, describing a course of argumentative reasoning that is followed to settle a case. Churning through large volumes of legal documents is a daily requirement for a large number of professionals who need access to the information embedded in them. Natural language processing methods that help in document summarization with key information components, insight extraction and question answering play a crucial role in legal text processing. Most of the existing document analysis systems use supervised machine learning, which require large volumes of annotated training data for every different application and are expensive to build. In this paper we propose a legal text analytics pipeline using Large Language Models (LLM), which can work with little or no training data. For document summarization, we propose an iterative pipeline using retrieval augmented generation to ensure that the generated text remains contextually relevant. For question answering, we propose a novel ontology-driven ensemble approach similar to cross-examination that exploits questioning and verification principles. A knowledge graph, created with the extracted information, stores the key entities and relationships reflecting the repository content structure. A new dataset is created with Indian court documents related to bail applications for cases filed under Protection of Children from Sexual Offences (POCSO) Act, 2012 an Indian law to protect children from sexual abuse and offences. Analysis of insights extracted from the answers reveal patterns of crime and social conditions leading to those crimes, which are important inputs for social scientists as well as legal system.