How harish-nesrekar/vyom works
Project type and purpose
• From the README (“VYOM: Enhancing OpenAI’s GPT-OSS with Multimodal Vision Capabilities extensible to ISRO EO Data”) and the mix of Next.js, Express/Python scripts, this is a full-stack, AI-powered web application.
• Frontend: Next.js-based UI for uploading images, viewing maps, chat interface, result cards.
• Backend: Node.js/Express API server (server.js, aiController.js, nasaController.js, authcontroller.js), plus Python-based microservices for vision/EO analysis (blip_loader.py, eo_analysis.py, vqa.py, risk_analysis.py, etc.).
File structure
• Top-level UI components (AuthContext.js, ChatBox.js, EOMapViewer.js, NavBar.js, etc.) and global styling (globals.css).
• frontend/ folder: Next.js app directory (app/page.js, app/layout.js, public assets), ESLint, PostCSS configs.
• Top-level Node code: server.js, api.js, ai.js, auth.js, authMiddleware.js – core Express routes, middleware, and API orchestration.
• Python scripts at root: modular microservices for loading models, running inference, generating captions, doing EO analysis, training (train_eurosat.py), evaluation.
• backend/ folder: legacy or additional Node modules (unclear – may house shared utilities or a separate microservice).
README for additional insights
• Extensibility to ISRO EO Data suggests pluggable loader components (e.g., nasa.js, nasaController.js, eurosat_loader.py).