{"slug": "how-i-built-an-ai-decision-copilot-to-help-india-prepare-for-the-2026-el-nino", "title": "How I Built an AI Decision Copilot to Help India Prepare for the 2026 El Niño Crisis", "summary": "A developer built an AI decision copilot to help India prepare for the 2026 El Niño crisis. The platform, developed during the Google Cloud Gen AI Academy APAC Hackathon, uses Gemini, Vertex AI, BigQuery, and Google Cloud to provide explainable recommendations for district administrators and farmers. The system integrates multiple public datasets and employs specialized AI agents for risk assessment, resource allocation, and personalized crop advice.", "body_md": "Building an explainable AI platform that helps district administrators allocate resources and farmers make better crop decisions using Gemini, Vertex AI, BigQuery, and Google Cloud.\n\nClimate disasters are not just weather events. They are decision problems.\n\nWhen forecasts predict a strong El Niño, governments do not simply need more data. They need answers to questions like:\n\nExisting dashboards provide plenty of charts.\n\nVery few provide decisions.\n\nThat became the motivation behind **El Niño 2026 Decision Copilot**, an AI-powered decision intelligence platform built during the **Google Cloud Gen AI Academy APAC Hackathon**.\n\nIndia depends heavily on the monsoon.\n\nA severe El Niño can lead to:\n\nThe information already exists across dozens of government portals, weather services, satellite datasets, and agricultural reports.\n\nThe challenge is that it is scattered.\n\nDistrict collectors do not have time to manually combine:\n\nFarmers face an even bigger challenge.\n\nMost need a simple answer:\n\nGiven my district, should I plant the usual crop this season?\n\nInstead of building another dashboard, I wanted to build an AI system that reasons over multiple data sources and produces explainable recommendations.\n\nThe platform serves two audiences through the same intelligence engine.\n\nThey receive:\n\nInstead of simply showing that a district has high risk, the system explains **why**.\n\nFarmers interact with a conversational AI.\n\nThey can ask questions such as:\n\nShould I grow paddy this season?\n\nor\n\nWhich crop is safer in my district?\n\nThe assistant retrieves official contingency plans, combines them with weather forecasts, and generates localized recommendations backed by citations.\n\nRather than building one giant AI model, I designed the platform as a decision pipeline.\n\n```\nData Sources\n      ↓\nData Ingestion\n      ↓\nBigQuery\n      ↓\nRisk Model\n      ↓\nAI Agents\n      ↓\nExplainable Decisions\n```\n\nEverything flows through a single decision intelligence core.\n\nThe platform runs entirely on Google Cloud.\n\nMultiple public datasets are collected continuously, including:\n\nAll datasets are standardized before being stored in BigQuery.\n\nBigQuery stores district-level information including:\n\nEvery downstream component reads from BigQuery instead of directly querying external APIs.\n\nThis keeps the architecture modular, scalable, and reproducible.\n\nThe first version uses **BigQuery ML**.\n\nInstead of training a black-box neural network, I intentionally chose an interpretable approach.\n\nThe model combines features such as:\n\nThe output is a district-level risk score.\n\nThat score becomes the starting point for the AI agents.\n\nA single prompt quickly becomes difficult to maintain.\n\nInstead, I divided responsibilities into specialized agents using Google's **Agent Development Kit (ADK)**.\n\nIts responsibility is to answer:\n\nWhich districts require attention first?\n\nIt ranks districts and explains the factors influencing each score.\n\nOnce priorities are known, this agent decides how available resources should be distributed.\n\nExamples include:\n\nUnlike pure LLM reasoning, resource allocation is deterministic so recommendations remain consistent and auditable.\n\nThis agent combines:\n\nFarmers receive localized advice supported by official documents instead of hallucinated responses.\n\nThe original hackathon prototype focused on 23 high-risk districts.\n\nScaling nationwide turned out to be far more challenging than building the AI itself.\n\nThe platform now covers:\n\nExpanding reliable public datasets was significantly harder than writing prompts.\n\nOne principle guided every feature.\n\nNever ask users to trust the AI blindly.\n\nEvery recommendation answers:\n\nFor government decision-making, transparency is often more valuable than model complexity.\n\nEvery source uses different schemas, update frequencies, and district identifiers.\n\nA significant portion of the project involved cleaning, validating, and reconciling datasets before any AI processing could begin.\n\nGovernment users need confidence in every recommendation.\n\nEvery output had to include evidence rather than relying on confidence scores.\n\nDistrict contingency plans vary significantly in formatting and quality.\n\nBuilding an automated indexing pipeline was necessary to move beyond a small proof of concept.\n\nThere are several areas I would continue developing:\n\nBuilding AI for public-sector decision making is very different from building a chatbot.\n\nThe hardest problems were not prompt engineering.\n\nThey were:\n\nLarge language models become far more useful when they serve as the reasoning layer instead of the entire application.\n\n🌐 **Live Demo**\n\n[https://climate-resilience-in.web.app](https://climate-resilience-in.web.app)\n\n🎥 **Demo Video**\n\nIf you are building AI systems for climate resilience, agriculture, disaster management, or public-sector decision intelligence, I would love to hear your thoughts and feedback.\n\nHappy building! 🚀", "url": "https://wpnews.pro/news/how-i-built-an-ai-decision-copilot-to-help-india-prepare-for-the-2026-el-nino", "canonical_source": "https://dev.to/naazim_hussain_7fbe686d11/how-i-built-an-ai-decision-copilot-to-help-india-prepare-for-the-2026-el-nino-crisis-2h1g", "published_at": "2026-07-10 06:36:56+00:00", "updated_at": "2026-07-10 06:41:51.687306+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "ai-agents", "ai-products", "ai-infrastructure"], "entities": ["Google Cloud", "Gemini", "Vertex AI", "BigQuery", "BigQuery ML", "Agent Development Kit (ADK)", "India"], "alternates": {"html": "https://wpnews.pro/news/how-i-built-an-ai-decision-copilot-to-help-india-prepare-for-the-2026-el-nino", "markdown": "https://wpnews.pro/news/how-i-built-an-ai-decision-copilot-to-help-india-prepare-for-the-2026-el-nino.md", "text": "https://wpnews.pro/news/how-i-built-an-ai-decision-copilot-to-help-india-prepare-for-the-2026-el-nino.txt", "jsonld": "https://wpnews.pro/news/how-i-built-an-ai-decision-copilot-to-help-india-prepare-for-the-2026-el-nino.jsonld"}}