1430. Building a RAG pipeline from scratch
Invited abstract in session WB-34: Theory of Knowledge, Technology, and Innovation, stream Advancements of OR-analytics in statistics, machine learning and data science.
Wednesday, 10:30-12:00Room: Michael Sadler LG10
Authors (first author is the speaker)
| 1. | Hisham Taj
|
| Analysis Hub, National Audit Office | |
| 2. | Alice Saunders
|
| Analysis Hub, National Audit Office |
Abstract
Retrieval augmented generation (RAG) pipelines are a popular method for resolving common shortfalls with large language models (LLMs). Whilst LLMs provide valuable assistance in generating text, summarizing information, and enhancing productivity, they also come with limitations, particularly the risk of hallucinations, where LLMs generate incorrect or misleading information with apparent confidence.
RAG pipelines offer a structured approach to mitigating these issues by retrieving relevant, fact-checked information from trusted sources before generating a response. This reduces the likelihood of hallucinations and improves the accuracy and reliability of the output.
Standard approaches to RAG pipelines typically rely on a wide variety of different libraries in Python that provide support for the creation of Vector stores and retrievers. Outside of Python however, there is limited support for these objects, meaning these structures often have to be created manually. For applications where the use of Python is not possible, this issue acts as a significant barrier to the implementation of RAG.
This talk will give an overview of how to create these structures from scratch, discussing not only the creation, but the optimisation of these structures, in order to give a full description of how RAG pipelines function.
Keywords
- Artificial Intelligence
- Machine Learning
- Analytics and Data Science
Status: accepted
Back to the list of papers