🌍 Inspiration We are running out of arable topsoil. The future of food security relies on "regenerative agriculture"—using the natural microbiome of the soil to feed plants instead of chemical fertilizers. However, there is a massive technology gap: analyzing soil DNA requires intense bioinformatics and high-performance computing (HPC) that is completely inaccessible to the average farmer.
We were inspired by the idea of being a "digital bridge." We wanted to build a tool that takes a massive, complex genomic file (like a 50GB FASTQ sequence) and translates it into something as simple as a video game health bar. Our goal was to prove that HPC doesn't have to be scary; it can be the invisible engine behind a simpler, greener future.
🛠️ How We Built It TerraScan is designed as a three-tier architecture that abstracts complexity away from the user:
The Frontend (Edge): Built with React Native (Expo), we created a clean, tactile interface where users can drag-and-drop files or simply type their location. We focused on clear visualization—using progress bars and "health scores" to represent complex genomic data.
The Backend (Orchestrator): We built a Node.js & Express server that acts as the job scheduler. It manages the queue, handles file uploads, and maintains a real-time connection to the app using Socket.io.
The "HPC" Engine: We developed a simulation engine that mimics the lifecycle of a distributed computing job. It models node allocation, parallel sequence assembly, and data merging, giving the user transparency into the "black box" of supercomputing without requiring them to write a single line of code.
The Agentic AI: We integrated Google’s Gemini 2.5 Flash model to act as our "Virtual Agronomist." Instead of hard-coded recommendations, the AI analyzes the simulated nitrogen and biodiversity metrics in real-time to generate context-aware farming advice specific to the user's region.
🧩 Challenges We Faced The "Zero Quota" Roadblock: One of our biggest technical hurdles was integrating the cutting-edge Gemini 2.0 models. We faced repeated 429 Too Many Requests errors due to API limits on experimental models. To solve this, we pivoted to the stable Gemini 2.5 Flash model and architected a "Bulletproof Fallback" system. If the AI service hangs or hits a limit, our backend seamlessly switches to a high-fidelity simulation mode, ensuring the user experience never breaks.
Bridging Localhost to Mobile: Connecting a physical mobile device to a local backend server was tricky due to network isolation. We had to implement dynamic IP configuration in our API service to allow the Expo Go app to communicate reliably with our Node.js cluster simulation.
Abstracting Complexity: It was difficult to decide how much "technical info" to show. We iterated on the UI to show just enough—like "Allocating 200 Nodes"—to make the user feel the power of the system, without overwhelming them with logs.
📚 What We Learned This project taught us that User Experience (UX) is the most important part of HPC. Powerful compute is useless if the people who need it can't access it. We learned how to use Generative AI not just for text, but as a data interpretation layer—turning raw numbers into human logic. Most importantly, we learned that building resilient systems means planning for failure; our hybrid AI/Simulation architecture made our final prototype robust and demo-ready.
Log in or sign up for Devpost to join the conversation.