Agentic RAG
- Planner/Superviser decides whether to retreive or browse or both
- Grader filters weak chunks
- Grounding Gaurd rejects answers that aren’t well supported and forces another loop/tool
- Binding tools to llm
- Refer Here for tools
- Refer Here for the code written
Deployment Considerations
- RAG:
- Raw Documents
- Vector Database
- Langchain/Langgraph application
RAW Documents
- Most of the organizations place these raw documents in blob storages
- AWS: S3
- Azure: Blob Storage
- GCP: google cloud storage
- Vector databases:
- Vector database as a service:
- PG Vector: This is add on to postgres
- RAG:
- Docker (Container)
- Docker will be deployed into k8s or any other container orchestration
