RGo.ai: A Web-Grounded, Schema-Constrained Agricultural Intelligence Platform with Vision-Language and Conversational LLM Integration

This paper has presented RGo.ai, a web-grounded, schema-constrained agricultural intelligence platform that integrates live marketplace signal ingestion, vision-language image analysis, domain-constrained conversational AI, and structured condition-based recommendation generation within a unified Next.js App Router application. The system demonstrates that lightweight real-time web signal extraction combined with schema-level output constraints and source-binding validation can serve as a practical, infrastructure-efficient alternative to heavyweight RAG architectures for domain-specific advisory applications. The four-modality architecture addresses distinct dimensions of the agricultural advisory gap: immediate input procurement guidance through grounded marketplace recommendations, rapid visual diagnosis of field conditions through multimodal image analysis, conversational access to agronomic knowledge through a domain-scoped LLM assistant, and structured, step-decomposed remediation guidance through the condition recommendation pipeline. Schema constraints and source-binding validation address the hallucination challenge at the output level without requiring embedding infrastructure or vector database deployment. Future development priorities include replacement of the heuristic anchor extractor with a more robust parsing approach resilient to HTML structure variation, implementation of a persistent signal cache with configurable time-to-live semantics, support for structured multimodal image attachments consistent with vision model API conventions, integration of per-IP rate limiting and authentication enforcement for production deployment, and conduct of the proposed human evaluation study to provide empirical grounding for the system’s relevance and actionability claims.  

[1] FAO, “The State of Food and Agriculture 2022: Leveraging Automation in Agriculture,” Food and Agriculture Organization of the United Nations, Rome, 2022. [2] T. B. Brown et al., “Language Models are Few-Shot Learners,” Advances in Neural Information Processing Systems (NeurIPS), vol. 33, pp. 1877–1901, 2020. [3] Z. Ji et al., “Survey of Hallucination in Natural Language Generation,” ACM Computing Surveys, vol. 55, no. 12, pp. 1–38, 2023. [4] V. Khamparia and B. Pandey, “A Novel Deep Learning-Based Approach for the Agricultural Advisory System,” Sustainable Computing: Informatics and Systems, vol. 22, pp. 316–323, 2019. [5] A. Kamilaris and F. X. Prenafeta-Boldú, “Deep Learning in Agriculture: A Survey,” Computers and Electronics in Agriculture, vol. 147, pp. 70–90, 2018. [6] S. P. Mohanty, D. P. Hughes, and M. Salathe, “Using Deep Learning for Image-Based Plant Disease Detection,” Frontiers in Plant Science, vol. 7, p. 1419, 2016. [7] H. Liu et al., “Visual Instruction Tuning,” Advances in Neural Information Processing Systems (NeurIPS), vol. 36, 2024. [8] P. Lewis et al., “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks,” Advances in Neural Information Processing Systems (NeurIPS), vol. 33, pp. 9459–9474, 2020. [9] R. Nakano et al., “WebGPT: Browser-Assisted Question-Answering with Human Feedback,” arXiv preprint arXiv:2112.09332, 2021. [10] L. Geng et al., “Grammar-Constrained Decoding for Structured NLP Tasks without Finetuning,” in Proc. Conf. Empirical Methods in Natural Language Processing (EMNLP), 2023, pp. 10932–10952.