NyayaSakhi–SWATI: India’s First Statute-Aligned, Retrieval-Augmented LAMP² 4.0 AI-Powered Digital Legal Companion for Victims of Domestic-Violence

Abstract

Domestic violence (DV) remains a pervasive challenge in India, and survivors frequently struggle to obtain timely, comprehensible, and statute-accurate information about their rights under the Protection of Women from Domestic Violence Act, 2005 (PWDVA). Existing helplines and portals are typically static, FAQ-based, or rule-driven, offering neither interactive legal literacy nor case-specific guidance grounded in statutory relief structures. This paper presents NyayaSakhi–SWATI, India’s first statute-aligned, retrieval-augmented AI legal companion for DV survivors, powered by the LAMP² 4.0 backend. SWATI (Support and Welfare Assistance through Technology Interface) is designed as an empathetic “AI NyayaSakhi” that securely collects survivor narratives, retrieves PWDVA-aligned precedents from the NyayaSmriti vector knowledge base, and delivers plain-language responses on likely statutory reliefs, indicative case duration, and prescriptive next steps. At the core of the system is the Statute- Aligned Legal Relief Prediction (SALRP) task, wherein LAMP² 4.0 predicts multi-relief outcomes (protection, residence, custody, monetary relief, and compensation) directly from narrative inputs. On the NyayaSmriti evaluation set, LAMP² 4.0 achieves an accuracy of 81%, outperforming strong Indian legal-AI baselines including INLegalLlama, PredEx, and NyayaRAG by margins of approximately +8–9 percentage points and +13.93 percentage points respectively. A single-case reference evaluation against an adjudicated PWDVA matter demonstrates complete alignment between predicted and judicially granted reliefs. A research preview study with domestic-violence survivors and experienced PWDVA practitioners further indicates that NyayaSakhi–SWATI is perceived as usable, empathetic, and statute-faithful. Together, these findings show how retrieval-augmented, statute-aligned AI can move beyond judgment-centric prediction toward survivor centric, pre-adjudicative decision support in domestic-violence litigation.