This little Amazon Bedrock feature is such a big deal! 😲
Yesterday, a colleague mentioned that their customer was struggling with trying out different models, because they had to re-write code every time they wanted to test a different model. 😩
“Did they use the new Converse API?”, I asked? 🤔
“What is it?”, they answered. 😕
“OMG your customer HAS to check this out”!, I exclaimed. 💡
Announced about three weeks ago, the Amazon Bedrock Converse API allows developers to code against a single, simple, straightforward, consistent API, then seamlessly switch between Models. 😎
Write once, chat with anything! 🎉
This is Amazon’s “Invent and Simplify” Leadership Principle at its best, making it easier for customers to test, choose, and replace the models they want, without the pain of having to rewrite code:
* Got your generative AI app working? Great! Can you optimize it for cost with a cheaper model? 💰
* Does your app work somehow, but quality could be better? Try a different model! ⬆️
* Want to offer free and premium versions of your app, by switching models? Just change a single attribute in your Converse JSON! 🔑
It’s not quite perfect yet: Different models may respond better to different system prompts, or require some other different settings. ⚠️
But fear not! The Bedrock Converse API has you covered: Since it’s JSON-based, all you need to do is update any attributes that need to change per model, including the system prompt. For example, you can set up a simple set of "plug-in adapter" JSON objects that reflect the differences between how you want to use each model, and use them to overwrite those attributes that change. 🔧
The Converse API also supports tool use, custom model attributes, and any of the programming languages supported by the AWS SDK, so check it out! 👀
What cool generative AI apps will YOU build with Bedrock and the Converse API? Share your ideas in the comments! 💬
Professional Services Director | General Manager | People Leader | Customer Champion | Business Builder
1yReally enjoyed your presentation Rachel Rui Liu - is there anyway for the attendees to access the content?