You're reading from Enterprise GENERATIVE AI Well-Architected Framework & Patterns

Product type Book

Published in Apr 2024

Publisher Packt

ISBN-13 9781836202912

Pages 113 pages

Edition 1st Edition

Languages

Concepts

Artificial Intelligence

Author (1):

Suvoraj Biswas

Table of Contents (43) Chapters

1. What is Generative AI?

2. What is the background of GenAI - Well Architected Framework?

3. Why do we need a Well Architected Framework for GenAI based Enterprise Use Case?

4. What are the pillars of the Well Architected Framework?

5. What are the building blocks for each pillar?

Operational Excellence:

6. What is a Large Language Model?

7. What is a Foundation Model?

8. When should we be using the Foundation Model?

9. Introduction to Content Moderation building block:

10. What is AI Guardrails and How to integrate it?

11. Introduction to the Compliance Pillar:

12. Benefits of having an Archival for Compliance building block:

13. What are some solutions for Archiving AI & Human interactions for Regulatory Compliance?

14. The FM or LLM observability Building Block:

15. What are some LLM Observability Solutions?

16. Introduction

17. What are some best practices to optimize Cost during Architecting?

18. Prompt Engineering

19. Why Prompt Engineering plays a crucial role in the GenAI apps?

20. Some Best Practices for Effective Prompts:

21. Embedding & Vector Database

22. What are Embeddings and why are they so important?

23. What are OpenAI's Offerings on Embeddings?

24. What is a Vector Database & How do they really work?

25. Options of Vector Databases for Solutioning

26. The Orchestrator Building Block

27. Some Orchestrator Application framework for AI development:

28. Introduction

29. What is Retrieval-augmented generation (RAG) Pattern for GenAI?

30. Demo Time - a GenAI based QnA python app using RAG pattern (LangChain, PineCone & Open AI api)

31. What is Fine tuning?

32. Why do we need fine tuning? Can’t we use RAG for domain specific use cases?

33. Are Fine Tuning & Pre-Training referring to the similar process?

34. Foundation Model or Large Language Model fine-tuning techniques:

35. Parameter Efficient Fine Tuning:

36. What are some Enterprise Use cases where SFT/RLHF or PEFT can be used?

37. What is LoRA (Low Rank Adaptation Model) technique?

38. What are the advantages of the LoRA method in fine tuning LLM?

39. What is QLoRA (Quantized Low Rank Adaptation Model) ?

40. Vector Database & Semantic Search capabilities

41. How to run PostGreSQL with pgVector extension in EC2 ?

42. Overview of AWS Sagemaker and AWS Bedrock

43. Example Generative AI based Enterprise Product Search Architecture in AWS

The Reliability pillar within our well-architected framework ensures the establishment of a highly dependable LLMOps based AI system, empowered by excellent platform level visibility, end to end traceability and system level performances. Additionally, any platform-level issues and concerns are actively monitored, prioritized and swiftly addressed by the responsible SRE, DevOps, or Engineering team. All involved parties should agree to a predefined SLA to guarantee a highly accessible LLM platform for the principal stakeholders.

The FM or LLM observability Building Block:

The LLM Observability platform is a crucial component that ensures the reliability of your system for users. This observability layer continuously monitors the ML Ops system from end to end, detecting any issues related to ML clusters, model drift, hallucinations, problematic prompts, prompt injection by end users, and...