SWATI AI : Advancing AI Models from Rule-Based Frameworks to NLP-Driven Prescriptive Analytics to Assist Victims of Domestic Abuse

SWATI AI (Support and Welfare Assistance through Technology Interface) is an AI-driven chatbot designed to assist victims of domestic abuse by providing accessible, non-judgmental, and actionable legal guidance. This paper introduces the LAMP² (Legal Analytics Model for Prediction and Prescription), the core framework that powers SWATI, progressing through four distinct research phases aimed at enhancing decision-making capabilities for victims of domestic violence under the Protection of Women from Domestic Violence Act, 2005. Phase 0 focuses on the creation of a curated dataset, PROTECT-JC, which compiles judgments from district courts, high courts, and the Supreme Court of India between 2008 and 2024. This dataset, developed in collaboration with legal experts and judicial professionals, serves as the foundation for training models to predict legal outcomes. Phase 1 introduces a rule-based decision framework in LAMP² 1.0, designed to extract and map predefined intents and entities from user inputs. This phase enables the structured documentation of complaints, providing victims with initial guidance based on the legal provisions of the Domestic Violence Act, 2005. The rule-based approach achieved accuracy of 80%, but with limited customization for individual cases. In Phase 2, LAMP² 2.0 leverages supervised machine learning models to predict legal outcomes, such as protection orders, custody orders, and monetary relief. These models, including Random Forest, XGBoost, and Decision Trees, achieved impressive results, with the Random Forest model showing 96.5% accuracy in predicting protection orders. Phase 3 evolves the model further with NLP-driven predictive analytics to process unstructured legal narratives, such as case judgments, transforming them into structured data for improved predictions. Models like Llama and Mixtral demonstrated strong summarization capabilities and show promise in predicting legal results. The integration of NLP and machine learning techniques allows for the prediction of case durations and other legal outcomes, addressing the complexities of the legal system. Overall, this research presents a comprehensive framework for assisting victims of domestic violence, combining rule-based models, machine learning, and natural language processing to offer predictive and prescriptive legal insights. The LAMP²-powered SWATI AI has the potential to significantly enhance access to legal assistance, empowering victims to make informed decisions regarding their legal options. This paper demonstrates the transformative potential of AI in providing personalized, data-driven legal guidance and advancing the accessibility of justice for underserved communities.