JusticeNetBD: Context-Aware AI to Enhance Legal Information Access for Bangladeshi Women via Retrieval-Augmented Generation

Women in Bangladesh often face barriers in accessing justice due to gender-based violence, workplace harassment, and limited awareness of legal protections. Despite existing laws, complex language and fragmented resources hinder women from seeking redress. Hence, many women do not realize a specific law exists for the harassment they faced, so they just feel uncomfortable, not knowing it was actually illegal. This paper proposes JusticeNetBD, a Retrieval-Augmented Gen- eration (RAG)-based legal assistant designed to provide accurate, context-aware, and accessible legal guidance tailored for Bangladeshi women. While the current focus is women’s legal rights, the framework is designed for gradual expansion to all Bangladeshi law domains via periodic corpus updates. Quantitative evaluations demonstrate that JusticeNetBD processes queries in 1–2 seconds on average, 10 times faster than ChatGPT-4o Turbo and DeepSeek-V3, and 2–5 times faster than Gemini Flash 2.5. Statistical analysis (Kruskal-Wallis H = 33.90, p < 0.001) confirms these differences are significant, with Dunn’s post-hoc tests revealing JusticeNetBD’s superiority (p < 0.0001). The system also excels in retrieval accuracy (Recall@2 = 0.90, MRR = 0.90) and answer quality (ROUGE-L = 0.463, BERTScore F1 = 0.896), outperforming general- purpose LLMs by 23–25 percentage points in ROUGE-L. By combining low-latency inference with grounded legal responses, JusticeNetBD offers a scalable solution for real-time legal aid in resource-constrained settings. The beta version of the application is hosted via Streamlit. Application access link and reproducibility guidelines are available in the GitHub repository: https://github.com/SakibHasanSimanto/JusticeNetBD.