RAG in Production — Phase 1
The Data Backbone

Source to searchable index. Every branch below is a scalability / resilience / cost decision, not a fixed recommendation — trace the path that matches your document volume and update frequency.

SCALABILITY RESILIENCE COST
Read every branch point in the diagram as a trade-off across these three forces — not as "the right way."
1

Data Ingestion

source → trigger → raw store
2

Parsing & Extraction

route by file type → extract → classify → human-in-the-loop
3

Preprocessing & Chunking

clean → chunk() interface → tag
4

Embedding

adapter → batch / retry / rate-limit
5

Storage & Indexing

cost vs. maintenance decision
Ingestion
Parsing / Extraction
Chunking
Embedding
Storage / Indexing
⋯ dashed = conditional path
◆ diamond = decision point
Hover any box for its role. Moving dots trace the direction of data flow through each layer.