Infrastructure Recommendations

Deployment guidelines for hosting the AI Dashboard application across cloud environments.

---

1. Stateless API Node Sizing

The FastAPI gateway handles compute-heavy tasks like schema serialization, Pydantic validations, and real-time SSE streaming generation.

Sizing Baseline

• CPU: Since multi-stage agent workflows run within localized Python async execution loops, assign multi-core instances (minimum 2 vCPUs) to prevent stream bottlenecks.
• Memory: Base workers need sufficient memory buffers (minimum 2GB RAM per running container) to process intermediate Pandas DataFrames efficiently.

---

2. Horizontal Scaling Guidelines

Because generation streams emit directly over open network connections, routing needs careful infrastructure configuration:

graph TD
    Client[Client Connection] --> LB[Load Balancer]

    subgraph "Stateless Application Nodes"
        LB --> Node1[FastAPI Gateway Instance 1]
        LB --> Node2[FastAPI Gateway Instance 2]
    end

    Node1 --> DB[(Shared MongoDB Cluster)]
    Node2 --> DB

Routing Considerations

• Session Restoration: Because final dashboard configurations are persisted centrally inside MongoDB, subsequent refinement operations (POST /dashboard/refine) can route to any healthy stateless gateway instance safely.
• Connection Throughput: Ensure downstream load balancers have generous request timeouts configured (e.g., 30+ seconds) to prevent dropping open SSE streaming sockets during complex agent generations.