I want to build a multi-tier application that replicates the core idea of DocsGPT while remaining fully self-hosted. The stack is already chosen: • Front end: ReactJS or AngularJS (your choice, but keep component structure clean and reusable) • Back end: Python with Django REST framework for the API layer • Database: MongoDB for persistent storage Vector handling The system must index and retrieve information through three separate numerical vector spaces. Each layer should be pluggable so I can experiment with different embedding models later without rewriting the pipeline. Core AI features 1. Text generation 2. Summarization 3. Question answering These should all draw on the vector stores and run on an on-prem LLM—no external API calls once the model is downloaded. A lightweight admin console is needed to upload new documents, trigger re-indexing, and monitor resource usage. User roles • Admin – full configuration, user management, and model maintenance • Editor – import documents, launch re-index jobs, view analytics • Viewer – search, ask questions, read generated summaries Scope & deliverables – Functional prototype wired end-to-end, deployable with Docker-Compose – Clean, documented React or Angular UI with role-based access control – Django API covering auth, vector operations, and inference endpoints – MongoDB schema plus initial seed data – Setup guide explaining model download, environment variables, and scaling tips – Brief test suite or Postman collection to verify the main flows I prefer clear, commented code and commit history that shows your thought process. Suggest additional best-practice improvements if you see gaps—I’m open to ideas as long as the above essentials are met.