reez-code/vunaguide-backend

AI-Powered Crop Doctor for Smallholder Farmers in Kenya 🇰🇪 # 🌿 VunaGuide Backend (API & Agents) Backend service for VunaGuide: An AI-powered agricultural assistant for Kenyan farmers. View Frontend Repository · Report Bug --- ## 📚 Project Deliverables Quick links to the Capstone Project submission materials: | Document | Link | | -------------------------- | -------------------------------------------------------------------------------------------------------------------------- | | 📄 **Project Proposal** | View Google Doc | | 🎥 **Demo Video** | Watch on Drive | | 📽️ **Presentation Slides** | View Google Slides | --- ## 📑 Table of Contents - Overview - Architecture - Tech Stack - Setup Guide - Deployment - Testing the API - License --- ## 📖 Overview This backend hosts the core Agentic logic for VunaGuide, orchestrating Google's Gemini 2.5 Flash models. It is responsible for diagnosing crop diseases from images and providing real-time, grounded farming advice via chat. --- ## 🏗️ Architecture (Google ADK) This service implements a **Manager-Worker Agent Pattern** using the Google Agent Development Kit (ADK) framework: ### 1. Manager Service (The Orchestrator) - Acts as the central router. - Decides whether to run the Visual Diagnosis Pipeline (if an image is present) or the Chat/Search Pipeline (if text only). - Enforces strict logic flow to prevent hallucinations. ### 2. Agronomist Agent (Worker) - A specialized `LlmAgent` equipped with vision capabilities. - Identifies plants, detects diseases (e.g., Maize Lethal Necrosis), and suggests remedies. ### 3. Sentinel Agent (Evaluator) - A dedicated safety auditor. - Reviews the Agronomist's output before it reaches the user to catch dangerous advice or banned chemical suggestions. --- ## 🛠️ Tech Stack | Category | Technology | Why we chose it …