Before the Petition: A Statute-Aligned Domestic Violence Legal Relief Prediction System in India

Dipali Awasekar

Keywords: Statute-aligned relief prediction, Domestic violence (PWDVA) analytics, NyayaDeepa legal corpus

TL;DR: An AI system that predicts legal reliefs for domestic violence victims in India—before they go to court—by aligning case details directly with the law.

Abstract:

Domestic violence proceedings are among the most urgent civil
matters brought before courts in India, yet they remain
plagued by delays, limited access to timely legal guidance, and
uncertainty around the statutory remedies realistically
available to victims. With Sections 18–22 of the Protection of
Women from Domestic Violence Act, 2005 (PWDVA) covering
protection orders, residence orders, monetary relief, custody
orders, and compensation, the absence of early, statutegrounded decision support often forces survivors to navigate
filing and settlement choices with incomplete information,
thereby amplifying legal risk, cost, and procedural burden. In
this paper, we present Before the Petition: A Statute-Aligned
Domestic Violence Relief Prediction System in India (IDVRPS),
an AI-powered framework designed to assist victims in prelitigation case investment decision-making by predicting
statute-aligned relief outcomes and generating legally
grounded explanations and prescriptive guidance based on
factual case attributes and statutory provisions. We curate and
release a comprehensive domestic-violence legal corpus from
NyayaDeepa, including a gold-standard curated subset and a
retrieval-ready knowledge base (NyayaSmriti) for RAG-based
statutory grounding. We develop a RAG-LegalTuned modeling
pipeline and evaluate its performance across multiple
configurations, benchmarking against four widely used Indian
legal AI baselines spanning legal summarization, legal QA and
reasoning, fact extraction with judgment prediction, and RAGbased label classification. Our results demonstrate that the
LLaMA 3.1–80B Legal-Tuned with RAG configuration
significantly outperforms the baselines, achieving ROUGE-1 =
0.512, ROUGE-L = 0.412, BLEU = 0.520, and Accuracy ≈ 81%,
with the highest lexical and semantic precision among
evaluated variants. IDVRPS offers a transparent, scalable, and
reproducible solution to support data-driven legal assistance
for domestic violence victims, improve early-stage remedy
awareness under PWDVA Sections 18–22, and establish a
research-grade benchmark for future statute-aligned legal AI
in India.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 4