RAG — Production Architecture
Production-grade RAG with hybrid retrieval, reranker, guardrails, and eval feedback loop.
Rendering…
Make it your own.
{
"nodes": [
{ "id":"u","type":"user","label":"User","x":40,"y":180 },
{ "id":"g1","type":"hexagon","label":"Input\nguardrail","x":200,"y":160,"fill":"#fee2e2" },
{ "id":"rw","type":"rounded","label":"Query rewrite","x":380,"y":160,"fill":"#dbeafe" },
{ "id":"bm","type":"database","label":"BM25 index","x":560,"y":40 },
{ "id":"vec","type":"database","label":"Vector DB","x":560,"y":160 },
{ "id":"kg","type":"database","label":"KG cache","x":560,"y":280 },
{ "id":"rerank","type":"rounded","label":"Reranker\n(cross-encoder)","x":760,"y":160,"fill":"#ede9fe","bold":true },
{ "id":"prompt","type":"rounded","label":"Prompt assembly","x":940,"y":160,"fill":"#dbeafe" },
{ "id":"llm","type":"aws-lambda","label":"LLM","x":1120,"y":160,"bold":true },
{ "id":"g2","type":"hexagon","label":"Output\nguardrail","x":1280,"y":160,"fill":"#fee2e2" },
{ "id":"out","type":"rounded","label":"Answer +\ncitations","x":1440,"y":160,"fill":"#dcfce7","bold":true },
{ "id":"tel","type":"aws-cloudfront","label":"Telemetry","x":1120,"y":320 },
{ "id":"eval","type":"aws-sns","label":"Eval harness","x":940,"y":320 }
],
"edges": [
{ "id":"a","source":"u","target":"g1","thickness":2 },
{ "id":"b","source":"g1","target":"rw" },
{ "id":"c","source":"rw","target":"bm" },
{ "id":"d","source":"rw","target":"vec","thickness":2 },
{ "id":"e","source":"rw","target":"kg" },
{ "id":"f","source":"bm","target":"rerank" },
{ "id":"g","source":"vec","target":"rerank","thickness":2 },
{ "id":"h","source":"kg","target":"rerank" },
{ "id":"i","source":"rerank","target":"prompt","thickness":2 },
{ "id":"j","source":"prompt","target":"llm","thickness":2 },
{ "id":"k","source":"llm","target":"g2","thickness":2 },
{ "id":"l","source":"g2","target":"out","thickness":2 },
{ "id":"m","source":"out","target":"tel","style":"dashed" },
{ "id":"n","source":"tel","target":"eval","style":"dashed" },
{ "id":"o","source":"eval","target":"rerank","style":"dashed","label":"retraining","curve":"step" }
]
}