KisanNama—Multi-Agent AI Crop Advisor for Pakistan

Pakistan's agriculture sector faces a recurring challenge: farmers often make planting decisions without access to reliable data about market demand, cultivation trends, or future risks. This frequently results in crop oversupply, falling prices, and major financial losses. KisanNama is an AMD-powered, multi-agent AI platform that helps Pakistani farmers choose the right crop before planting. By combining agricultural data, market trends, and large language models, it delivers personalized recommendations that reduce risk and improve profits. The platform uses four AI agents, each with a specific role. The Data Agent retrieves district crop data, the Risk Agent estimates oversupply risk from year-over-year planting trends, the Market Agent analyzes historical prices, and the Strategy Agent recommends suitable and alternative crops. Together, they provide practical, data-driven advice tailored to each farmer. To improve accessibility, KisanNama generates responses in Urdu and English at the same time, letting farmers use their preferred language. Recommendations stream in real time using Server-Sent Events (SSE), while text-to-speech converts AI responses into natural speech for users with limited literacy. Built with FastAPI, CrewAI, Gemma 4 31B on a self-hosted vLLM server running on AMD cloud GPUs, and PostgreSQL, KisanNama showcases how agentic AI can solve real agricultural problems. The platform also includes JWT authentication, chat history, and an admin dashboard for managing crops, districts, cultivation records, and market prices. Our goal is to give farmers timely, data-driven insights that reduce planting risk, increase income, and promote sustainable farming. By bringing advanced AI to local languages, KisanNama bridges the gap between modern technology and Pakistan's farming community