Multimodal RAG Platform · View 03 of 16
Layered Architecture
Six layers with dependencies pointing downward only. Establishes what may call what.
Copy
PNG
PDF
⋯
Editable source
SVG
draw.io
All views
Experience
Experience
Web App
React + TypeScript
Web App...
Chat & Upload
Chat & Upload
Source Viewer
page / frame highlight
Source Viewer...
Admin Console
Admin Console
API
API
API Gateway
OIDC, quota
API Gateway...
Query API
FastAPI, SSE
Query API...
Ingestion API
Ingestion API
Admin API
Admin API
Orchestration
Orchestration
Query Orchestrator
plan, retrieve, answer
Query Orchestrator...
Ingestion Workflows
Temporal, durable
Ingestion Workflows...
Model Gateway
route, cache, cap
Model Gateway...
AI services
AI services
Layout & OCR
Layout & OCR
Vision Captioning
Vision Captioning
Transcription
Transcription
Embedding
text + image
Embedding...
Reranker
Reranker
Guardrails
Guardrails
Data
Data
Vector Index
Vector Index
Lexical Index
Lexical Index
Analytical Tables
Analytical Tables
Object Store
Object Store
Metadata DB
Metadata DB
Platform
Platform
Identity & Policy
Identity & Policy
Secrets & KMS
Secrets & KMS
Observability
Observability
Audit & Lineage
Audit & Lineage
HTTPS
HTTPS
in-process
in-process
gRPC
gRPC
search / upsert
search / upsert
Layered Architecture
Layered Architecture
Application we own
Application we own
Interface / broker
Interface / broker
Data store
Data store
Security / platform
Security / platform
synchronous
synchronous
Dependencies point downward only. No layer calls the layer above it.
Dependencies point downward only. No layer calls the layer above it.
v 1.0 · owner Solution Architecture
v 1.0 · owner Solution Architecture
Text is not SVG - cannot display
Layers
Experience — web app, chat, source viewer, admin
API — gateway plus query, ingestion and admin surfaces
Orchestration — query orchestrator, workflows, model gateway
AI services, Data, Platform
Rules
No layer calls the layer above it
Only the model gateway may reach a model provider
Only AI services and orchestration may reach the data layer
Why it matters
Model swaps stay inside the gateway
Index changes stay inside the data layer
Each layer is separately testable and separately scalable
