Copyright © 2026 Authors retain the copyright of this article. This article 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.

@article{209662,
        author = {Dr MK Jayanthi Kannan and Ankur Singh},
        title = {AI Lawyer: A Literature Review of Retrieval-Augmented Legal Advice, Document Automation and Forensic Document Verification},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {5},
        pages = {4826-4835},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=209662},
        abstract = {Legal information is spread across long statutes, rules, and case material; routine legal documents are drafted by repeating standard text, and digital documents submitted as evidence may have been edited without notice. Large language models (LLMs) can answer legal questions fluently, but they can also produce confident, incorrect statements, which is serious in a legal setting. This paper reviews the literature behind AI Lawyer, a Python and FastAPI prototype that brings three functions behind one API: retrieval-augmented legal question answering, template-based generation of a non-disclosure agreement (NDA), and preliminary forensic screening of PDF and image files. Eight studies on retrieval-augmented generation, sentence embeddings, legal language models, legal hallucination, image forensics, and local language models are compared on objective, technology, methodology, efficiency, and open issues. 
The comparison shows that retrieval, domain adaptation and forensic heuristics are each well studied on their own, while their integration in a private, locally deployable legal workflow is not. The paper then describes the proposed system, its algorithms and parameters, and reports a small preliminary test of the Error Level Analysis (ELA) module on synthetic images: the default threshold of 35.0 flagged none of 60 images, and the score separated edited from authentic images only weakly (AUC 0.61). The legal knowledge base has not yet been populated, so answer quality is not claimed. The paper closes with the contribution, limitations, and a plan for evaluation.},
        keywords = {Retrieval-Augmented Generation, Legal Informatics, Large Language Models, Sentence Embeddings, Document Generation, Digital Forensics, Error Level Analysis, FastAPI.},
        month = {October},
        }