Summarize by Aili
Learnings in Q1 ’24 about Knowledge Graphs and RAG
🌈 Abstract
The article discusses the potential of knowledge graphs and structured knowledge representation within existing Retrieval Augmented Generation (RAG) pipelines. It highlights the growing market for RAG, the need for more granular tooling for structured knowledge representation, and the concept of "minimum viable graphs" as the future of GraphRAG.
🙋 Q&A
[01] Why the market for RAG is huge
- Most white-collar work involves information retrieval, as workflows can be described as retrieving information, transforming it, performing basic reasoning, and providing an answer or report.
- Information retrieval can be documented as a workflow, and structured knowledge representation can improve information retrieval.
- RAG currently faces issues with information retrieval due to the lack of granular tooling for structured knowledge representation, which companies like WhyHow.AI are actively building.
- There is a need for workflow tooling that allows non-technical domain experts and RAG/Agent developers to collaborate and translate business workflows into RAG systems.
[02] RAG is a process, not a system
- Each step of the RAG process is nuanced and reflects the individual complexity of the workflow.
- The goal is to have more modular tools for granular control and context injection, working alongside more intelligent reasoning foundation models.
- RAG systems that claim to work out of the box can only do so for undifferentiated information workflows, which may not be suitable for more complex use cases.
[03] Minimum Viable Graphs as the future of GraphRAG
- The concept of "minimum viable graphs" suggests that knowledge graphs can be used to sharpen semantic focus, rather than just as aggregators of data.
- Small, use-case-specific graphs can be sufficient, as LLMs can add their own understanding on top of the structured data.
- This means that it is not always necessary to create a perfectly described version of the world, as the LLM can fill in the gaps.
[04] WhyHow.AI's KG tools in closed beta
- Deterministic Document structure-based retrieval
- Knowledge Graph SDK
WhyHow.AI is building tools to help developers bring more determinism and control to their RAG pipelines using graph structures.
Shared by Daniel Chen ·
© 2024 NewMotor Inc.