The latest articles on DEV Community by Madhumitha R (@madhumitha_r_a284a9d152cb). https://dev.to/madhumitha_r_a284a9d152cb https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F2741594%2F02316723-e6f9-4377-a0f4-3824d580d69e.png https://dev.to/madhumitha_r_a284a9d152cb en Madhumitha R Mon, 31 Aug 2026 08:29:35 +0000 https://dev.to/madhumitha_r_a284a9d152cb/how-i-built-rescuroute-ai-2ib5 https://dev.to/madhumitha_r_a284a9d152cb/how-i-built-rescuroute-ai-2ib5 <h1> RescuRoute AI </h1> <p>I created <strong>RescuRoute AI for this hackathon</strong> to explore how agentic AI can help rescue perishable shipments when route disruptions and temperature failures threaten delivery.</p> <h2> The Problem </h2> <p>Perishable shipments can become difficult to save when unexpected route incidents or refrigeration failures increase delivery time and reduce remaining shelf life.</p> <p>The goal of RescuRoute AI is to continuously evaluate the shipment situation and recommend the best available action instead of simply following a fixed route.</p> <h2> How I Built It </h2> <p>RescuRoute AI uses an event-driven architecture built with Google Cloud and Google ADK.</p> <p>The workflow is:</p> <p><strong>Shipment Event → Google Pub/Sub → Cloud Run → RescuRoute Agent → Decision Tools → Rescue Decision → Firestore</strong></p> <p>The agent uses <strong>Google ADK and Gemini</strong> for reasoning and orchestration. Gemini does not perform the deterministic route calculation or spoilage calculation itself. Instead, the agent reasons over structured evidence produced by specialized deterministic tools.</p> <h2> Decision Intelligence </h2> <p>The system evaluates:</p> <ul> <li>Route travel time and incident delays</li> <li>Remaining product shelf life</li> <li>Temperature abuse and spoilage risk</li> <li>Alternate route feasibility</li> <li>Nearby buyer options when delivery is no longer viable</li> </ul> <p>The decision can change as conditions worsen.</p> <p>For example:</p> <p><strong>Normal conditions → MONITOR</strong></p> <p><strong>Critical temperature + viable alternate route → REROUTE</strong></p> <p><strong>No safe route within remaining shelf life → COMMERCIAL_RESCUE</strong></p> <h2> Google Cloud Deployment </h2> <p>The production workflow is deployed using:</p> <ul> <li>Google Pub/Sub for event delivery</li> <li>Google Cloud Run for the HTTP processing endpoint</li> <li>Firestore for persistent rescue decisions and audit records</li> </ul> <p>The deployed Cloud Run service successfully received an HTTP <strong>200</strong> request during validation.</p> <h2> Testing </h2> <p>The project includes automated unit tests covering the agent workflow, route logic, spoilage logic, buyer evaluation, rescue decisions, and persistence.</p> <p>The final test run completed with:</p> <p><strong>46 passed</strong></p> <h2> What I Demonstrated </h2> <p>The final demo shows the complete flow from a shipment event to an intelligent rescue decision, including:</p> <ol> <li>Normal shipment monitoring</li> <li>Route incident handling</li> <li>Critical temperature/spoilage risk</li> <li>Automatic rerouting</li> <li>Commercial rescue when routing is no longer safe</li> <li>Cloud Run deployment</li> <li>Firestore persistence</li> <li>Automated test validation</li> </ol> <h2> Conclusion </h2> <p>RescuRoute AI demonstrates an end-to-end agentic workflow for making time-sensitive logistics rescue decisions using structured evidence, Google ADK, Gemini, and Google Cloud.</p> <p><strong>This project was created specifically for this hackathon.</strong></p> <h3> Project Repository </h3> <p><a href="https://github.com/madhu-17-SE/rescuroute-ai" rel="noopener noreferrer">https://github.com/madhu-17-SE/rescuroute-ai</a></p> ai