RAG Q and A with Bedrock and Astra - Who died in Romeo and Juliet?
summary
Builds a working RAG pipeline that loads 321 lines (one line per document) from Act 5 Scene 3 into a vector store and answers “who died” using embeddings and an LLM. Uses Amazon Bedrock (Titan embeddings for vectors, then Anthropic Claude for completion) with DataStax AstraDB as a real‑time Cassandra vector database, LangChain/Casio glue, and a Jupyter notebook. Demo succeeds; model choice changes answers, and internal tests claim up to 18x (and ~4x) performance gains versus competitors.
AI-written summary · about
chapters (15)
- 0:00RAG with Bedrock and Astra
- 1:46Amazon Bedrock and Astra
- 3:59Vector databases in RAG
- 6:49Building RAG demo steps
- 8:56Python environment setup
- 11:55Astra and AWS credentials
- 13:12Vector store for Shakespeare
- 15:51Prompt design for answers
- 17:17RAG query and results
- 18:41Q&A Astra and Cassandra 5.0
- 20:50Astra DB benefits
- 24:33Astra DB vs vector DBs
- 26:00Vector and non-vector tables
- 28:18RAG accuracy challenges
- 29:23Closing and resources
related talks
53mBuild RAG from Scratch
20mProduction RAG apps made easier with Astra
23mNo RAGrets
39mWhat if you could build production-ready RAG entirely in Elixir?
41mFrom RAGs to riches: Build an AI document interrogation app in 30 mins
28mBuilding database queries with Amazon Bedrock
14mVector Embeddings and RAG Demystified
56mRAG: from dumb implementation to serious results