Skip to content

Deployment Topology (Canonical)

Overview of environments, network boundaries, and Azure service mapping for the MVP and near-term scale.

Environments

Env Purpose Differences
dev Fast iteration, feature branches Lower SKU, no cost-critical alerts
staging Pre-release validation Same infra templates as prod, reduced scale
prod Customer traffic Autoscale + budget alerts + stricter policies

Azure Service Map (MVP)

Layer Service Notes
API / Backend App Service (Linux B1) Consolidated API + minimal SSR
Frontend Static Web Apps CDN edge + auth integration option
Ingestion Functions (Consumption) Timer & HTTP triggers
Storage (Raw) Blob Storage Versioned raw payloads (immutable)
Relational Data Azure SQL (Basic) Trips, profiles, normalized petals
Search / Vector Cognitive Search (Basic) Hybrid queries (BM25 + vector)
ML Inference Azure ML Managed Endpoint or ACI Embeddings + ranking prototype
Secrets / Identity Entra ID + Key Vault (future) Start with managed identities only
Observability App Insights / Log Analytics Centralized logs + metrics + traces

Network & Security (Early)

  • Public endpoints acceptable MVP; plan private endpoints + VNet integration pre GA.
  • Managed Identity for intra-service auth (no stored secrets).
  • CORS restricted to primary domain + localhost dev.

Scaling Path

Pressure Signal Action
Search query latency p95 > target Raise Cognitive Search replica count
Embedding queue backlog > threshold Add worker (Functions concurrency)
Model cost / req rising Introduce smaller re-ranker + distillation
Cache miss ratio > planned Add Redis tier (Basic) + semantic response cache
SQL DTU saturation Move to S0 / evaluate Postgres flex for geospatial

IaC Notes

  • Use Bicep modules per layer (api, ingestion, data, ml, monitoring).
  • Parameterize SKU & region; enforce cost ceiling via policy.

Last updated: September 2025