Advanced RAG Applications with Vector Databases
1h 18mAdvanced2024-10-17
Authors
Yujian Tang
Course details
Retrieval-augmented generation (RAG) is everywhere these days, and vector databases are what give them their power. But RAG isn’t as simple as some companies claim, so it can be easy to get overwhelmed. In this course, discover state-of-the-art RAG methods, including how to optimize text-based RAG via chunking, embedding, and metadata usage, and how to conduct basic image search with a vector database. You’ll also get a chance to practice multimodal RAG by embedding and storing data and querying images with text. Along the way, instructor Yujian Tang provides practical, hands-on demonstrations and exercise challenges to test out your new skills.
Skills covered
Hugging FaceNatural Language Processing (NLP)Machine LearningDatabase DevelopmentPythonDatabase ManagementArtificial Intelligence (AI)Programming LanguagesOpen SourceSoftware DevelopmentOne-Off
Concepts
Introduction
- RAG with vector databases - Advanced strategies for AI optimization
- What you should know
- Setting up your exercise files
Optimizing RAG
- Introduction to preprocessing for RAG
- Chunking considerations
- Chunking examples
- Introduction to embeddings
- Embedding examples
- Metadata
- Demo - Chunking
- Demo - Metadata
- Demo - Embed and store
- Demo - Querying
- Demo - Adding the LLM
- Challenge - Cite your document sources
- Solution - Cite your document sources
- Challenge - Change the chunk size
- Solution - Change the chunk size
Image Search with Vector Databases
- Introduction to vector embeddings for images
- Vision models 101
- Demo - Getting semantic vectors
- Demo - Storing image vectors
- Demo - Comparing images semantically
- Challenge - Find the dog most similar to a cat
- Solution - Find the dog most similar to a cat
Multimodal RAG with Vector Databases
- Introduction to the types of multimodality
- Ways to do multimodal RAG
- Introduction to multimodal embedding models
- Demo - Embedding and storing data
- Demo - Query images with text
- Challenge - Find anomalies in your embeddings
- Solution - Find anomalies in your embeddings
Conclusion
- Next steps