How CDL Handles Security and Reliability of their GenAI Workloads with Amazon Bedrock.
Join us to discover how CDL tackles emerging risks in GenAI workloads. Through live demonstrations, learn practical strategies for testing, securing, and monitoring GenAI workloads using Amazon Bedrock.
The episode featured Matt Houghton from CDL, an insurtech company in the UK, discussing how they handle security and reliability of their generative AI workloads on Amazon Bedrock. Matt, who is both an architect at CDL and an AWS Community Builder/Ambassador, explained how CDL is using AI to build digital insurance journeys and make their APIs AI-ready. He shared his extensive experience speaking at various AWS events and conferences, highlighting his passion for data and community involvement.
The main technical discussion centered around CDL's implementation of AI guardrails and evaluation processes. Matt demonstrated how they use Amazon Bedrock to create and test AI models, showing both automated and human evaluation processes to ensure accuracy and safety. He explained their approach to testing different models against specific metrics like helpfulness, correctness, and faithfulness, while also considering responsible AI metrics such as stereotyping and harmfulness. The presentation included live demos of their guardrail system and how it handles both text and image-based content.
The final portion of the show focused on how CDL is using Bedrock's data automation features to transform unstructured data into structured formats. Matt demonstrated this by showing how their system could analyze a receipt image and extract relevant information into useful formats like CSV and markdown, making it ready for further processing in data pipelines. Throughout the presentation, Matt emphasized the importance of continuous evaluation and monitoring of AI systems, especially as new models become available.
Key Highlights:
- CDL is implementing AI to enhance insurance customer journeys
- They use both automated and human evaluation processes for AI models
- Guardrails are implemented to prevent harmful or inappropriate AI responses
- The system can handle both text and image-based content
- Data automation features help convert unstructured data to structured formats
- Continuous evaluation and monitoring are essential for AI system maintenance
Hosts of the show
Tony Mullen - Senior Database Specialist @AWS
Guests
Matthew Houghton - Data Architect @CDL
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article