Welcome! Here you'll find answers to some of the most commonly asked questions.
Q: Can I provide just the raw vectors to the LLM in order to save storage on the HANA database excluding metadata and chunk information?
A: You can't give an LLM raw vectors (like arrays of floats) and expect the LLM to "understand" or decode them back into natural language. Embedding vectors are lossy and directional — they’re designed for similarity comparison, not reversible interpretation.
Q: Where do I put the contextual information in a classic RAG flow?
A: Put the retrieved text chunks into the user message, not the system message.
- The system message is best used to set overall behavior, tone, or identity of the assistant (e.g., “You are a helpful technical assistant.”).
- The user message is where the contextual information should go — this allows the LLM to consider it as part of the question, giving it full attention when formulating the response.
Q: Where can I learn more about AI? A: You can take a look at the Further Learning sections of each exercise and in the main Readme of this repository.
Below is the direct link to the Further Learning section of the main Readme:
Learning section in the main README
Below are direct links to the Further Reading sections in each exercise:
- Exercise 00 - Set up your workspace
- Exercise 01 - Explore Generative AI Hub in SAP AI Launchpad
- Exercise 02 - Create the SAP AI Core connection configuration
- Exercise 03 - Explore the SAP HANA Cloud Vector Engine
- Exercise 04 - Create the database schema for the CAP application
- Exercise 05 - Create vector embeddings using an embedding model
- Exercise 06 - Define the Job Posting Service
- Exercise 07 - Implement the Job Posting Service
- Exercise 08 - Understand data masking, anonymization and pseudonymization
Q: How can I get in touch?
A: Send me a message on LinkedIn or via mail