Loading article...
Loading article...
Stream/Area:
-
Subject:
-
Paper ID:
IJSREM68668
Volume:
10 |
11 |
Nov 2026
Pages:
1-6
Publication Frequency:
Monthly (12 issues per year)
License:International Journal of Scientific Research in Engineering and Management follows the Creative Commons Attribution 4.0 (CC BY 4.0) License, allowing free use and distribution with proper attribution.
Department of Computer Engineering
Vishwakarma Institute of Technology, Pune
Department of Computer Engineering
Vishwakarma Institute of Technology, Pune
Department of Computer Engineering
Vishwakarma Institute of Technology, Pune
Department of Computer Engineering
Vishwakarma Institute of Technology, Pune
Department of Computer Engineering
Vishwakarma Institute of Technology, Pune
In India, legal data is still extremely unequally accessible because the majority of citizens cannot interpret the Bharatiya Nyaya Sanhita (BNS) along with the multitude of other jurisdictions without paying for professional legal help. In this paper, we introduce the full-stack AI platform that attempts VidhiSetu to alleviate this problem using three groundbreaking innovations: (i) deep multilingual legal generation, which forces Google Gemini 1.5 Flash to generate legal advice, fine amounts, and news reports in a user-specified language by structuring the prompts and returning validated JSON output without losing the nuances of law in automated translations; (ii) spatio-legal inference, which resolves the jurisdictional intersection of Central, State, and Municipal legislations based on a user's GPS coordinates; and (iii) hallucination-free generation, which uses the enforcement of strict JSON Pydantic schema at the level of Python FastAPI to prevent any hallucinations of statutory references. Additionally, the system supports multi-modal search queries (textual, audio, or document upload), interactive legal knowledge graphs automatically generated with AI, live Indian legal news aggregation with automatic severity classification by AI, SOS dispatcher and location-based tracking, and verified lawyer directory by geography. Evaluation performed using a curated dataset of 50 BNS-related legal queries results in an accuracy of 86% and an average latency of 1.34 seconds. VidhiSetu shows that careful prompt engineering and architecture design can turn an LLM into a legal intelligence system with domain constraints and language flexibility.
KEYWORDS
In this paper, we have introduced VidhiSetu, a multi-modal location-aware legal informatics solution built to solve the accessibility issues associated with the Indian legal system. By employing an asynchronous FastAPI gateway alongside strict Pydantic type-guards, our platform ensures structural determinism of raw token streams outputted by an LLM, thereby reducing statutory fabrication and hallucinations from 23.3% to a much safer rate of 4.0%. Additionally, since the language parameters are directly fed into the prompt layers, the platform attains an absolutely flawless 100.0% JSON structure success rate along with a boost of +18.0% for multilingual factual consistency for regional Indian languages such as Hindi and Marathi. The platform also incorporates real-time GPS streams via browsers along with dynamic visualization modules like react-force-graph-2d canvas elements, proving the case that generative AI can indeed be made deterministic and reliable for civic empowerment. Although the existing architecture is extremely stable in terms of providing real-time statutory routing capabilities, there are some architectural developments that are scheduled to be implemented in subsequent releases. First, to move away from the purely zero-shot prompt injection mechanism to absolute knowledge, the idea is to implement an enhanced RAG backend that relies on a vector database where raw and authenticated gazette notifications can be indexed for use with absolute knowledge routing. The second way to improve geospatial routing capabilities would be to substitute simplistic coordinate-based geolocation with advanced geofencing using polygons to account for complex overlapping zones of municipal governance as well as local corporate jurisdiction. And finally, in future releases, there will be the ability to incorporate a decentralized verification protocol for professional user states by implementing an immutable registry of regional legal advisors.
[1] UNDP India, “Access to Justice in India: Challenges and Opportunities,” United Nations Development Programme, New Delhi, India, 2022. [2] Office of the Registrar General, “Census of India 2011: Language Atlas,” Ministry of Home Affairs, Government of India, New Delhi, 2011. [3] H. Zhong, Z. Guo, C. Tu, C. Xiao, Z. Liu, and M. Sun, “Legal judgment prediction via topological learning,” in Proc. 2018 Conference on Empirical Methods in Natural Language Processing (EMNLP), Brussels, Belgium, pp. 3540–3549, 2018. [4] I. Chalkidis, M. Fergadiotis, P. Malakasiotis, N. Aletras, and I. Androutsopoulos, “LEGAL-BERT: The muppets straight out of law school,” in Proc. Findings of the Association for Computational Linguistics: EMNLP 2020, pp. 2898–2904, 2020. [5] A. Blair-Stanek, N. Holzenberger, and B. Van Durme, “GPT-4 passes the bar exam,” arXiv preprint arXiv:2212.14402, 2023. [6] J. Niklaus, I. Chalkidis, and M. Stürmer, “Swiss-judgment-prediction: A multilingual legal judgment prediction benchmark,” in Proc. Natural Legal Language Processing Workshop, EMNLP 2021, pp. 19–30, 2021. [7] N. Aggarwal, A. Malhotra, and P. Bhatt, “A Hindi legal question answering dataset and evaluation of multilingual pre-trained models,” in Proc. International Conference on Natural Language Processing (ICON), India, 2022. [8] [8] Z. Ji et al., “Survey of hallucination in natural language generation,” ACM Computing Surveys, vol. 55, no. 12, pp. 1–38, Mar. 2023. [9] P. Manakul, A. Liusie, and M. J. F. Gales, “SelfCheckGPT: Zero-resource black-box hallucination detection for generative large language models,” in Proc. 2023 Conference on Empirical Methods in Natural Language Processing, pp. 9004–9017, 2023. [10] K. Janowicz, P. Hitzler, B. Adams, D. Kolas, and C. Vardeman, “Five stars of linked data vocabulary use,” Semantic Web, vol. 5, no. 3, pp. 173–176, 2014.
Publication Date
11/7/2026
Peer Review Type
Double-Blind Peer Review
Review Rounds: 2
Documents
Article Views
0
Total views since publication
PDF Downloads
3
Total downloads since publication
Geographic Reach
5
Digital Preservation: This article is permanently archived on Zenodo and complies with all data preservation standards across major indexing platforms.
Plagiarism Screening
This manuscript has been checked for plagiarism using TurnitinPlagiarismChecker software.
Conflict of Interest
The authors declare no conflict of interest related to this research work.
Funding Information
No funding information available for this research.
Related research in the same subject area
Loading similar articles...