Cloud Architecture & Deployment Plan
| Components | Azure Service |
|---|---|
| Prototype | Hosted on AZ vm |
| Authenticated user login | Az appservice (Azure identity provider) |
| Frontend (Web APP) | Az app service |
| Backend API | Az Function |
| Document Storage | Az Blob Storage |
| Metadata / logs | Az cosmos DB |
| Vector Embeddings | Az AI Search |
| LLMs (Agents) | Az OpenAI Service |
| Monitoring and logs | Az Monitor + Az Application Insights |
Phase 1: Prototype Deployment
Initially, the entire application will be hosted on a single Azure Virtual Machine (VM). This allows rapid iteration and validation before modularizing services into dedicated Azure resources.
| Component | Deployment (Prototype Phase) |
|---|---|
| Full Stack Prototype | Azure Virtual Machine (Linux) |
| All services | Containerized/locally hosted on VM |
Phase 2: Modular Cloud Services Deployment (Post-Prototype)
Once validated, each component will be migrated to its respective managed Azure service for scalability, security, and cost efficiency.
Frontend (Document Upload + Chat Interface)
-
Service: Azure App Service (Web App)
-
Features:
- Scalable web hosting for UI and APIs
- Use Standard or Basic Tier for cost savings
-
Cost Optimization:
- Auto-scale during business hours only
- Use CDN (Azure Front Door or Azure CDN) for static assets
API Layer (Document Processing & Orchestration)
-
Service: Azure Functions (Serverless)
-
Use for:
- Upload handling
- API endpoints
- Triggering workflows
-
Cost Optimization:
- Pay-per-execution
- Best for variable workloads (spiky traffic)
NLP Processing Engine
- Service: Azure Machine Learning or Azure Container Instances
- Model Hosting Options:
Use Azure Kubernetes Service (AKS) with spot instances for LayoutLM- Use Azure OpenAI Service for GPT-based processing if available
Vector Search (RAG System)
- Service: Azure Cognitive Search + Azure Cosmos DB
- Details:
- Use Cosmos DB for metadata + hierarchical JSON storage
- Use Azure Search with vector index for semantic search
Storage Systems
- Service: Azure Blob Storage
- Usage:
- Raw documents
- Processed JSON
- Generated videos, audio, and images
- Cost Optimization:
- Use Hot tier for recent uploads, Cool/Archive for older assets
- Use lifecycle rules to auto-move blobs to cheaper tiers