Building on Bedrock

August 03, 2024

We started using Amazon Bedrock at work about a year ago to thoughtfully integrate large language models (LLMs) into some of our internal tools and processes. I want to write a series of posts capturing what I've learned from the effort and to riff on "what-if" alternatives in some places. I expect the result to wander through topics like:

• setting up for happy local development
• abstracting LLM use with the Bedrock Converse API
• letting a model call external functions with Bedrock Tool Use
• populating and querying Bedrock Knowledge Bases
• structuring larger Streamlit applications for extension and maintenance
• deploying Slack Bolt apps to a serverless platform
• single-table design with pynamodb / DynamoDB
• provisioning infrastructure with AWS CDK as a Terraform veteran
• observing how the system is working with AWS X-Ray as a Datadog adept

You won't go from zero-to-chatbot in 10 minutes reading what I write here. There are plenty of better resources for that. Rather, I hope this series highlights design considerations and ancillary work needed to use LLMs and realize their value.

I have no timeline for publishing, only this entry as a starting point.