Evidence-Grounded and Explainable Agentic Rag for Reliable Decision Support in Indian E-Governance
Main Article Content
Abstract
Government departments in India now publish a great deal of information online: schemes, eligibility rules, required documents, laws and administrative procedures. Citizens need answers they can trust and check, and AI tools are increasingly used to provide them. This study compares three ways of answering such queries: a standard large language model (LLM), conventional retrieval-augmented generation (RAG), and an evidence-grounded, explainable Agentic RAG framework. We tested all three on 250 Indian e-governance queries spread over five areas: government schemes and benefits, citizen services and documentation, laws and regulations, public welfare and eligibility, and administrative procedures. Answers were scored on accuracy, relevance of evidence, source attribution, completeness of context, hallucination, traceability, explanation and overall reliability. The Agentic RAG framework came out ahead of both alternatives, with 94.6% answer accuracy, 93.8% evidence relevance, 96.2% source attribution and 94.1% overall reliability. It also scored 93.4% on evidence traceability, 94.3% on evidence-to-claim consistency, 95.8% on hallucination-free responses and 94.0% on overall hallucination mitigation. Taken together, the results suggest that retrieving evidence in several passes, checking it, attributing sources and validating the final answer makes AI-generated decision support for e-governance more accurate, more transparent and easier to audit.
Article Details
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
References
Pandey, P., & Singh, J. P. (2026). Explainable Rating Prediction and Sentiment-Segmented Summarization for E-Governance Applications. SN Computer Science, 7(7), 715.
Papageorgiou, G., Sarlis, V., Maragoudakis, M., Magnisalis, I., & Tjortjis, C. (2025). Evaluating faithfulness in agentic RAG systems for e-governance applications using LLM-based judging frameworks. Big Data and Cognitive Computing, 9(12), 309.
Chennupati, P. K., Varanasi, V. S. V., Fakruddin, S., Yarlagadda, E. A., Devi, J. R., & Manasa, B. (2026, June). Explainable Customer Feedback Analysis Using Graph-Based Orchestration and Emotion-Aware LLM Reasoning. In 2026 5th OPJU International Technology Conference (OTCON) on Smart Computing for Innovation and Advancement in Industry 5.0 (pp. 1-6). IEEE.
Tavanxhiu, T., Godolja, M., Sevrani, K., & Naco, M. (2026). A technological blueprint for smart and AI-driven hospitality in emerging tourism markets: Evidence from Albania. Systems, 14(2), 188.
Hamza, A., Rizwan, M. B., Kayani, M., & Sheikh, S. A. A. (2026). Legalyze: An Open‐Source Structure‐Aware Agentic RAG Platform for High‐Precision Legal Intelligence. Expert Systems, 43(10), e70415.
Ganesan, C. (2026). A Policy-Aligned Agentic RAG Framework for Risk-Aware Decision Support in Enterprise Customer Relationship Management. Journal of Intelligent Decision Making and Information Science, 3(8s), 1409-1443.
Prabhu, R. S., Manikandan, S. P., Poudel, A., Maity, D., & Kumar, M. S. (2026). Generative AI Technologies for Intelligent Decision-Making Applications. International Journal of Artificial Intelligence and Machine Learning, 6(10s), 1530-1552.
Raj, A., Hareesh, B., Ramannavar, M., & Ankalkoti, P. (2026). Agent Breast-RAG: An Agentic Retrieval-Augmented Explainable Large Language Model Framework for Multimodal Breast Cancer Diagnosis and Clinical Decision Support. Computer and Decision Making: An International Journal, 3(3S), 01-13.
Wang, Z., Li, J., Guo, X., & Tsuchiya, T. (2026). From noisy feedback to evidence-aware issue specifications: an agent-governed retrieval-augmented generation approach. Automated Software Engineering, 33(3), 103.
Satpute, B., Rahane, W. P., Pawar, P., Vanjari, H., Kulkarni, R., Parhad, S. V., & Deshmukh, P. V. (2026). Multi agent retrieval validation and knowledge reasoning for enhanced retrieval augmented generation. Scientific Reports.
Gakhar, S., & Singh, R. K. (2026). Leveraging large language models (LLMs) for GeoAI-enabled digital agro-advisory. Frontiers in Remote Sensing, 7, 1839369.
Patil, S. S. (2026). Explainable Artificial Intelligence In Ayurvedic Clinical Decision Support For Prameha: Opportunities, Challenges And Future Directions. European Journal of Prosthodontics & Restorative Dentistry, 34, 148.
Arafin, P., Moniruzzaman, M., & Alkhateeb, A. (2026). Evidence-Grounded Clinical Pharmacogenomics Decision Support: Integrating Fine-Tuned Large Language Models with Hybrid Retrieval-Augmented Generation. Applied Sciences, 16(18), 9148.
Lakhani, T. L. (2026). From Static Recommendations to Autonomous Decisions: A Framework for Agentic Recommendation Systems with Memory, Reasoning and Tool Use. International Journal of Sustainability and Innovation in Engineering, 4(1).
Sablatnig, R. (2025). Retrieval-Augmented Generation for Enterprise Security Intelligence in Cloud-Based Decision Support and Knowledge Systems. International Journal of Research Publications in Engineering, Technology and Management (IJRPETM), 8(6), 13480-13488.
Wu, K. (2026). Trust-Calibrated Enterprise Service Copilots: Cost-Aware Agentic RAG for Customer-Support Knowledge Management. Stout in Business and Management, 2(1), 01-
Kafle, D. S., Talhi, E., & Coustaty, M. (2026, August). RAGXDoc: Structured Knowledge-Guided Retrieval and Explainable Re-ranking. In Document Analysis and Recognition–ICDAR 2026: 20th International Conference, Vienna, Austria, August 31–September 2, 2026, Proceedings, Part III (p. 70). Springer Nature.
Alghamdi, M. M., & Alamri, W. A. (2026). IKEA: Intelligent Knowledge Extraction with Reasonable and Agentic AI Agents. Applied Sciences, 16(15), 7704.
Huang, Y., Chen, Y., Zhang, H., Li, K., Zhou, H., Fang, M., ... & Wang, J. (2025). Deep research agents: A systematic examination and roadmap. arXiv preprint arXiv:2506.18096.