DigestAI news desk
hot Google DeepMind launches Gemini 3.8 Live and 3.8 Live Extended Thinking models new Critics label Big AI's proposed development slowdown a market lockout strategy hot TypeSafe AI unveils Jev, a frontier model up to 400× cheaper and 200× faster new Anthropic projects profitability for second consecutive quarter hot OpenAI uses hundreds of contractors to review ChatGPT chats new NepKANUN: RAG-based AI assistant improves access to Nepali legal information hot Researchers Introduce 'Never Give Up' to Fix RL Reasoning Plateaus in LLMs new Mapping Self-Reported Personality Archetypes of 22 Large Language Models hot OpenAI Unveils GPT‑6 Astra: Record‑Breaking 3D Rendering, Loop‑Transformer Architecture new Retrieval‑Augmented LLM Boosts Intersection Safety Recommendations from Crash Narratives hot OpenAI acquires AI camera software startup Glass Imaging for over $300 million new Comment on arXiv:2607.01233: Survivorship Bias Concerns hot Mistral AI raises €3B at €21B valuation to expand European sovereign AI infrastructure new Typos Disrupt Prompt‑Injection Probes in LLMs, Study Finds hot Google, OpenAI, Anthropic Discuss Industry AI Safety Standards Body new New framework optimizes LLM inference costs via adaptive model activation
Research updated

NepKANUN: RAG-based AI assistant improves access to Nepali legal information

Researchers have introduced NepKANUN, an AI-powered legal assistant designed to address the challenges of accessing legal information in Nepal. The system utilizes a fine-tuned large language model integrated with a Retrieval-Augmented Generation (RAG) framework to provide accurate answers to natural language legal queries. This approach aims to mitigate issues related to complex legal…

1 source primary source

Key points

  • NepKANUN is a RAG-based assistant using a fine-tuned LLM for Nepali legal texts.
  • The system achieved BERTScore F1 scores of 0.82, 0.77, and 0.71 for simple, moderate, and complex queries.
  • Expert reviews confirmed the tool's usability in providing precise answers to natural language legal inquiries.

The model was trained on a custom dataset consisting of high-quality question-answer pairs specific to Nepali legal texts. Performance evaluations using BERTScore indicate strong results, with F1 scores of 0.82 for simple queries, 0.77 for moderate ones, and 0.71 for complex legal questions. These metrics suggest that the system can effectively handle a range of difficulty levels in legal inquiries.

Beyond quantitative metrics, the usability of NepKANUN was validated through expert reviews, confirming its practical application. The study highlights the potential of combining generation and retrieval techniques to democratize access to legal knowledge. By focusing on customized local data, the project demonstrates a viable path for deploying specialized AI tools in low-resource linguistic contexts, potentially serving as a model for other regions facing similar barriers to legal information access.

Read the original at arXiv cs.CL · by Bhabuk Thapa, Prasiddha Koirala, Ranjit Raut, Sunil Regmi, Bal Krishna Bal primary source Open source ↗
Topics · follow one to build your own front page
NepKANUN

The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us.

Comments

via GitHub Discussions

More in Research

All →

Related stories