EPIC: RAG Part 2

[gphorvath]

Problem Statement:

One of the limitations of Large Language Models (LLMs) is that they are only able to respond to scenarios contained within training data - and training on new data is expensive given the size of the model. Retrieval Augmented Generation (RAG) is a technique to supplement the LLM with new data to enable it to provide more up-to-date responses.

Acceptance Criteria:

• IDAM for managing access to RAG Data
• API is compliant with OpenAI endpoints (Chat, Embeddings, Files, Assistants at a minimum).
• Handle GPT2 (bottleneck) in RAG Backend
• Better embeddings model (currently Instructor-XL)
• Smarter / Better RAG

Definition of Done:

Tasks

Preview Give feedback

1. ADR: RAG Refinement Approaches #267
   documentation +1
2. ADR VectorDB #272
   documentation +1
3. ADR for Text Embeddings #273
   documentation +1