What is RAG and Q for Business?

Retrieval Augmented Generation (RAG) is a natural language processing technique that enables generative AI models to reference an authoritative knowledge base before generating an answer.

RAG can be used to build AI assistants tailored to a specific business domain. Amazon Q for Business enables the rapid creation, tuning and deployment of RAG solutions.

Self-build RAG challenges:

Whilst tooling exists to enable organisations to build similar capabilities, self-build comes with challenges such as:

• Internal expertise in LLM architectures and frameworks needs to be developed.
• Appropriate hardware and software need to be purchased.
• On-going maintenance and operational capabilities need to be in place.
• Security and compliance requirements need to be met.
• Data management, versioning and re-indexing need to be managed to ensure ongoing accuracy of the assistant.

Who is it for?

Organisations with internal domain specific knowledge bases; for example, standard operating procedures and market intelligence. Example verticals include:

• Financial services such as Insurance, Banking, Investments, Capital Markets
• Life Sciences and manufacturing
• Energy

How it works:

1. Discovery: The RAG accelerator beings with an initial discovery workshop which typically takes one day. The aim of the workshop is to identify high business value use cases. Following the workshop, we determine the which of the identified use cases to pilot based on the availability of data, security and compliance requirements and end user needs. We also determine a set of objective business KPIs to measure during the pilot. In total this phase takes takes approximately one week.
2. Pilot: During the pilot phase, we configure and deploy the Q RAG application(s) along with the associated infrastructure required to securely connect to data sources and end users and to measure the identified KPIs. Deployment of the pilot typically takes two to three weeks. We then allow time for users to engage with the assistant to collect feedback.
3. Plan: Objective KPIs are measured and, along with subjective user feedback we evaluate the business impact of the pilot. Based on these measurements we determine a rollout plan to move to a full production implementation. The plan typically addresses issues such as data security, classification, versioning and role-based access.
4. Roll Out: Post pilot, we can support secure and compliant rollout of the RAG solution across the organization.

Benefits:

• Boost team productivity: Q for Business streamlines workflow by summarising documents, generating drafts, conducting research, or running comparative analysis.
• Dramatically reduced code and infrastructure: Compared to self-build, a Q implementation has a far lower implementation and maintenance overhead, reducing TCO.
• Sustainability: Using centrally trained and managed models, as opposed to self-training Q improves your company’s carbon footprint.

Why fourTheorem?

• We wrote the book on it: AI as a Service is a detailed guide to operationalizing Cloud Native AI Services on AWS.
• The AI-Powered Underwriting Assistant: Our Augmented Underwriting RAG assistant has been trained on 24 years of Lloyds of London public market bulletins.