Six developers share real-world Amazon Bedrock-powered apps

Tons of new large language models (or LLMs) are becoming available on Amazon Bedrock these days — from Claude 3 to Mistral to Llama 3. What’s cool about the playground that Bedrock offers is that you can accomplish so many different tasks using these various models together or separately.

Here are six creative applications built by developers in the last few months using various models on Amazon Bedrock.

Generalized chatbots like ChatGPT or Claude can be interesting, but they’re not all that reliable when you’re asking for highly specific or specialized information. To help with this problem, Senior Software Engineer Somil Gupta created a personalized assistant using Mistral Large on Amazon Bedrock to query over his collection of PDFs, so it could answer specific questions and provide reference material for verification. It’s easy to imagine researchers, medical professionals, or others with extensive personal notes using RAG in this way to create more personalized AI assistants, along with more verifiable responses.

One way clients let companies know what they need is through JIRA tickets — but while those can be helpful, handling them one-by-one doesn’t always address more persistent issues. AWS Community Builder Amelia Hough-Ross decided to put Claude 2 on Amazon Bedrock to work on the issue, both noting “what our customers needed from our team, and if there were any trends I could identify to provide better self-service and FAQ documentation.” Although the tool is still in development, Hough-Ross has been learning about the power of prompt engineering to refine the quality of results she gets from the LLMs.

Some companies enable direct communications between customers and sellers via a wide variety of channels, but platforms don’t want to empower bad actors to scam customers: that’s bad for people, and it’s bad for business. To help with this problem, Omid Eidivandi put together a simple application that used Amazon Bedrock to detect potentially fraudulent messages between customers on a fictitious website. Check out how he did it.

Unlike others I talked to for this post, Alan Blockley wasn’t building an application for work, but rather thinking about a practical use for LLMs in his life. He was particularly struck by “the overall impact that the whole Claude 3 family has on the LLM landscape,” and in particular “the claim of ‘strong vision capabilities.’” Blockley wanted to know how strong those capabilities really were. So he built a photo album application to harness “the poetic prowess of Claude 3 Haiku to provide insightful summaries of uploaded images.” Not only does this help with basic searchability, but Blockley observed that it “can create powerful use cases such as making art more accessible for the blind or improve operational overhead in managing catalogs of stock images in a creative studio.”

Solutions Architect Majdi Dhissi decided to build something a little different on Amazon Bedrock, too: an AI image generator that uses “a microservices approach to scale and have high availability via a clustered message broker.” Dhissi wrote an extensive blogpost about how he approached and built this project — and all the ways it could be implemented, scaled, and further developed in the future.

Finally, my colleague Banjo Obayomi has been playing with Amazon Bedrock for months, recently using it to have LLMs play each other in Street Fighter and Pokemon — or trying to have it beat Slay the Spire. While these experiments might not seem easily applicable to business use cases, I love reading about them partly because of the window they provide into how generative AI works — or doesn’t work.

All of these examples are exciting to read about, and they get at the experimental nature of many of these AI tools. If you want to read more about what people are building — or if you want to start learning to building your own experimental applications — check out the generative AI space on community.aws! And if you're building anything cool with Amazon Bedrock, let me know in the comments!