Ch 10 — RAG Solutions Landscape — Under the Hood
Internals of frameworks, vector DBs, platforms, and integration patterns
Under the Hood
-
Click play or press Space to begin...
ALangChain Architecture InternalsLCEL, Runnables, and the component model
1
input
Input
Dict or string
Runnable
1
link
LCEL Chain
pipe operator |
invoke
1
output
Output
Structured result
2
compare_arrowsLlamaIndex vs LangChain: index-centric vs chain-centric architecture
BVector Database Architecture ComparisonHow Pinecone, Qdrant, Weaviate, and pgvector differ under the hood
3
cloud
Pinecone
Serverless pods
vs
vs
vs
3
database
pgvector
PostgreSQL ext
4
analyticsBenchmarks: QPS, latency, recall at different scales (1M, 10M, 100M vectors)
CEmbedding Provider InternalsAPI design, batching, rate limits, and cost optimization
5
batch_prediction
Batch API
Efficient embedding
cache
5
cached
Local Cache
Avoid re-embedding
serve
6
attach_moneyCost analysis: API vs self-hosted embedding at different scales
DPlatform Integration PatternsHow Bedrock, Azure AI Search, and Vertex AI wire RAG together
7
cloud_upload
Ingest
S3 / Blob / GCS
chunk
7
auto_fix_high
Auto Pipeline
Managed chunking
query
7
smart_toy
Generate
With citations
8
securityEnterprise: IAM, encryption, VPC, compliance (SOC2, HIPAA, GDPR)
EEvaluation & Observability InternalsHow Ragas, LangSmith, and tracing work under the hood
9
science
Ragas Metrics
LLM-as-judge
trace
9
monitoring
LangSmith
Trace every step
alert
9
bug_report
Debug
Find failures
FMigration & Integration PatternsSwitching components, avoiding lock-in, and building for change
10
swap_horiz
Abstraction Layer
Swap components
migrate
10
sync
Re-embed
New model migration
test
10
verified
Validate
A/B test quality