Retrieval-Augmented Legal Document Summarization and Case Prediction using FAISS

Abstract

Due to the exponential rise in the number of digital legal documents such as court judgments, FIRs, contracts, and legal orders, it is becoming difficult and cumbersome to analyze them manually. In this regard, this paper proposes LawTech AI, which is a smart legal document analysis tool that helps transform unstructured legal documents into structured forms. In the proposed approach, multiple stages of processing begin with the preprocessing phase and Legal Named Entity Recognition (NER), followed by the FASSI processing flow (Fetch, Analyze, Summarize, Store, and Interact). This way, the contextual analysis of legal documents will be captured effectively through this workflow. To ensure fast searches within the document collection for conducting legal research and precedents, document embeddings are created and indexed in FAISS vector database. This helps in searching semantically similar legal case documents efficiently. Moreover, the suggested methodology involves employing a RAG model, where documents extracted from the legal corpus will be used to fine-tune a legal language model, thus enabling it to create structured summaries, detect legal issues, and gain insight.

Introduction

The text discusses the development of LawTech AI, an AI-based system designed to improve the analysis of large volumes of legal documents that are typically unstructured and difficult to search manually. Traditional legal research is time-consuming and inefficient, while existing AI approaches (such as Legal-BERT, machine learning classifiers, and retrieval models) often face limitations like poor interpretability, high computational cost, or inability to fully capture complex legal reasoning.

To address these issues, LawTech AI introduces an integrated pipeline combining multiple NLP and AI techniques. It uses Legal Named Entity Recognition (NER) to extract key legal entities such as case names, judges, courts, and legal provisions. It follows a structured FASSI workflow (Fetch, Analyze, Summarize, Store, Interact) to systematically process legal documents. The system also uses text embeddings stored in a FAISS vector database to enable semantic search for similar cases instead of simple keyword matching.

Additionally, a Retrieval-Augmented Generation (RAG) framework is used, where relevant cases are retrieved and provided as context to a fine-tuned language model (Mistral-7B) to generate structured summaries and insights. Sentence embeddings are generated using models like all-MiniLM-L6-v2.

The system is trained and evaluated using the Supreme Court of India judgment dataset (1950–2024) from Kaggle. Performance is measured using metrics such as ROUGE-1, ROUGE-L, cosine similarity, and Recall@K to evaluate summarization quality and retrieval accuracy.

Overall, LawTech AI aims to make legal research faster, more accurate, and more efficient by combining entity extraction, semantic search, and advanced language model-based summarization in a unified system.

Conclusion

In the present work, the possibility of using an AI-Powered Legal Document Intelligence System (LawTech AI) in the analysis of large groups of legal documents has been examined. The work has been specifically targeted at the transformation of unstructured legal texts into a more organized and queryable form through the application of modern Natural Language Processing techniques. The Legal Named Entity Recognition (NER) technique has been employed to recognize significant information such as the names of cases, courts, judges, legal acts, sections, and parties involved in the legal case. The proposed workflow included incorporating the FASSI approach, which facilitated the system in fetching, analyzing, summarizing, storing, and interacting with legal document data in a systematic manner. To facilitate the retrieval of cases, document embeddings were created and stored in a FAISS vector database, allowing for similarity-based document search in legal cases. Once a query or document is given, the relevant cases are fetched and used as context in a model known as Retrieval Augmented Generation (RAG), allowing the model to generate a concise summary and provide insights that enable a user to easily understand the key points in a case. The results of the present work indicate that the integration of entity extraction, semantic search, and AI-based summarization can greatly contribute to the analysis of legal documents. The proposed system can be helpful to lawyers, legal researchers, as well as law students in effectively dealing with large amounts of legal information. The reduction in the time required to analyze long legal documents can greatly contribute to the ease of access of legal information. As for suggestions for further work, it can be enhanced by adding more data to the database by using a larger number of legal documents from different legal cases and jurisdictions. The precision of entity recognition and reasoning can also be enhanced to improve the quality of generated summaries. Moreover, a user interface and legal search tools can be added to enable the system to be used more effectively. With further development, AI-based legal intelligence systems such as LawTech AI can play a significant role in assisting contemporary legal analysis and decision-making processes.

References

[1] D. Chalkidis, I. Androutsopoulos, and N. Aletras, “Neural Legal Judgment Prediction in English,” Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics (ACL), pp. 4317–4323, 2019. [2] N. Aletras, D. Tsarapatsanis, D. Preo?iuc-Pietro, and V. Lampos, “Predicting Judicial Decisions of the European Court of Human Rights: A Natural Language Processing Perspective,” PeerJ Computer Science, vol. 2, pp. 1–19, 2016. [3] I. Chalkidis, M. Fergadiotis, P. Malakasiotis, N. Aletras, and I. Androutsopoulos, “LEGAL-BERT: The Muppets Straight Out of Law School,” Findings of the Association for Computational Linguistics (EMNLP), pp. 2898–2904, 2020. [4] J. Zhong, C. Xiao, C. Tu, T. Zhang, Z. Liu, and M. Sun, “How Does NLP Benefit Legal System: A Summary of Legal Artificial Intelligence,” Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (ACL), pp. 5218–5230, 2020. [5] D. Lewis, J. Yang, T. Rose, and F. Li, “RCV1: A New Benchmark Collection for Text Categorization Research,” Journal of Machine Learning Research, vol. 5, pp. 361–397, 2004. [6] J. Devlin, M. Chang, K. Lee, and K. Toutanova, “BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding,” Proceedings of NAACL-HLT, pp. 4171–4186, 2019. [7] N. Reimers and I. Gurevych, “Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks,” Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing (EMNLP), pp. 3982–3992, 2019. [8] J. Johnson, M. Douze, and H. Jégou, “Billion-scale Similarity Search with FAISS,” IEEE Transactions on Big Data, vol. 7, no. 3, pp. 535–547, 2021. [9] P. Lewis, E. Perez, A. Piktus et al., “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks,” Advances in Neural Information Processing Systems (NeurIPS), vol. 33, pp. 9459–9474, 2020. [10] T. Brown et al., “Language Models are Few-Shot Learners,” Advances in Neural Information Processing Systems, vol. 33, pp. 1877–1901, 2020. [11] Y. Yang, Y. Cer, A. Ahmad, M. Guo, J. Law, and N. Constant, “Multilingual Universal Sentence Encoder for Semantic Retrieval,” Proceedings of ACL, pp. 87–94, 2019.

Copyright

Copyright © 2026 Mrs. T. Nagamani, M. Prachay kumar, G. Nani, N. Chaitanya, P. Pradeep Kumar. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.