Written by Matt Merriel
For those people who have been living on Mars for the last 10 years, with their eyes closed, and their fingers in their ears (10 points if you get the reference)… This thing called Generative AI can be taught to do things for you and make you look way smarter and more productive than you are. And 2024 was FULL of new services, products and announcements from our friends over at Amazon Web Services… There are so many in fact that it's hard to determine what everything does or what problems it solves. The beginning of a new year is a great opportunity to take a minute to review the current state of things and see which ones might be useful for solving problems in your organisation.
Two quick things to note before we dive in:
Given we are an Australian/New Zealand-based company, we'll be looking at things through the lens of an organization based in Australia. Some of the points raised may not be as relevant if you are located in other parts of the world.
Generative was NOT used to write this article however some Gen-AI-based services (such as Grammarly) were used to conduct a review once the article was written to improve the overall quality and readability of the article.
New Nova Models for Bedrock
One of the big announcements from Re: Invent 2024 was the launch of the new Amazon Nova generation of foundational models that can process text, images, and video as prompts. This opens up a lot of possibilities around how users can interact with the system and the knowledge/understanding that can be gained from it.
If your LinkedIn feed was anything like mine in early December you would have seen loads of posts from people generating images and Video using the other 2 Models that were announced:
Amazon Nova Canvas which is an image generation and editing model, and
Amazon Nova Reel which is targeted at generating video footage. (currently limited to generating 6 seconds of 1280x720 footage)
Based on the examples provided, these are quite capable and good candidates for producing content to complement/accompany other material (such as images for blog posts, marketing material etc.).
What's nice about Nova Canvas is that it's more than a simple Mid-Journey alternative… It can take an image as input and perform various actions on it including:
Background removal - Automatically identifies multiple objects in the input image and removes the background. Can even provide a transparent background as the response.
In-painting - The ability to reconstruct masked areas to remove/replace objects within an image
Image Variation - Provide example images and a prompt to generate images that preserve the content of the input images
Like most Generative AI models… All of this can already be done by talented people with the right skills, but for people like me who can barely spell Photoshop let alone use it… This would come in very handy.
While the Nova family of models are currently only available in a select number of regions, expect that number to grow throughout 2025. What's going to be even more exciting though, is the announcement that two additional models are looking to be released throughout 2025, including:
A speech-to-Speech model allows for new ways to interact with the model, and
MultiModal-to-MultiModal… meaning "any to any" interactions. This means it will be able to both input and output Audio, Video, Text and Images.
Add in the lower cost of the Nova models, and they deserve a look if you're planning on doing some generative-AI work this year.
Q Suite
The other Generative-AI suite that has gotten a lot of love over the last few months is the Amazon Q suite of products. These products offer a more standardized offering and interface… in return for a much lower, more "low/No Code" level of initial configuration.
Amazon Q Developer
I'll be honest, I was NOT a fan of Q Developer when it first came out in April of last year. GitHub Copilot had been out for a while and was doing everything I thought I needed quite well. At the same time, Q Developer was slower, typically gave lower-quality responses and always wanted me to log back into my Amazon Account.
But it's seen a LOT of additions since then and is now a competent, performant coding assistant that I'd use as my daily driver if only Emacs were a supported IDE. Since it was released it's gained the ability to generate:
Develop a new feature via a text prompt
Write Unit Tests
Perform Code Reviews
Generate Documentation
Across more than 25 different programming languages (the exact capabilities of language-to-language can be found in the user guide available here). If you haven't looked at it recently, I highly suggest you spend a few days with it… I'll follow up on this article with a more in-depth look at how it works.
As technologists, we’re used to things changing. Every year or two, there is a new technology/methodology for us to learn. But in the Artificial Intelligence space, that window is measured in weeks and months not years. Last month's leader can be next month's laggard, which means that to stay ahead, we must remain curious and re-evaluate our options. This is why we’ll release more articles on Generative-AI throughout 2025 to try and help cut through the noise and highlight the solutions and offerings that shine.
Amazon Q Business
To quote the official AWS documentation, "Amazon Q Business is a fully managed, generative-AI powered assistant that you can configure to answer questions, provide summaries, generate content and complete tasks based on your enterprise content, and complete tasks based on your enterprise data." That's a fancy way of saying, "It's like having a ChatGPT console, but based on your internal company data". It can scrape documents, knowledge bases, confluence sites… the works, and provides a fully managed website to interact with. Moreover, all the data that’s scrapped or produced by the Q Business is stored within your AWS account, allowing you to apply the required level of security and compliance controls… which is not available through some other SaaS offerings. What's more is that it fully supports Agent-based actions, meaning you can use it to automate tasks by leveraging AWS Lambda functions… all within a simple chat-bot style interface.
We use this ourselves internally at Mantel Group for several tasks such as surfacing data for reports/tenders and the like from multiple sources… as well as automating weekly tasks like report generation and calendar bookings. For the types of "wouldn't it be nice if we could just" problems… Amazon Q Business is a great first step. You can always take your Lambdas and guardrails and leverage them in Bedrock if you outgrow Q Business in the future or want some additional customisation options.
The elephant in the Room (or NOT in the room as the case may be)
Over the last 6-8 years, AWS has gotten a lot faster at distributing new services to regions outside of the big 3 (being North California, North Virginia and Ireland) to the point where more and more ap-southeast-2 launching regions in a lot of cases. This, however, does not appear to be true when it comes to services and features within the artificial intelligence space, and this results in a lot of potential opportunities that are just not available to organisations in the Asia Pacific area.
A quick look into Model Access at the time of writing (Jan 11th 2025) shows that ap-southeast-2 currently has 12 Models available in Bedrock whereas North Virginia has 47. And the recent introduction of the Model Catalogue with Bedrock Marketplace hasn't helped much either with 131 models available in the Catalogue in Sydney vs 169 models in the North Virginia Catalogue. Now a number of the models available in North Virginia are the older generations of the model that probably wouldn't make sense to launch in other regions given that newer more suitable models exist… But that's hardly the only difference.
The Sydney region is still missing some pretty common models including:
All of the AI21 Labs Jurassic and Jamba models
Claude 3.5 Haiku, 3.5 Sonnet V2 and Claude 3 Opus.
All of the Meta models
Stability AI's Stable Diffusion.
The ap-southeast-2 region doesn't have a single model capable of generating images.
Now, some might argue that in the big schema of things… it doesn't matter as you can call Bedrock in another region and use the models available there… And that's 100% correct, you can technically do that and it works surprisingly well and with remarkably low latency. Assuming your data is allowed to leave Australia and you have security, governance and compliance controls set up to support/enable/track/log all of that traffic/interactions in a different region. If you don't, or you can't… then your options are:
Wait for a suitable model to become available
Leverage SageMaker JumpStarts which allows you to run dozens of publicly available and proprietary models in your AWS account… including some of the ones not currently available through Bedrock. This can be an even better option as in some cases, it allows for greater fine-tuning of the model to provide a more optimised output.
All of this is to say that, at least for the time being… If you have data that must stay within Australia but want to leverage some of that sweet sweet Gen-AI goodness, then take some time to review your options before you start that next Proof of Concept… Or better yet, feel free to reach out to us. We'll help you navigate your options as we've done for numerous customers already (who don't like a good old call to action in a blog article).
Conclusion
To summarise, AWS has spent a lot of time, money and effort developing their AI product offering over the last 12-16 months, resulting in many options for customers to choose from. Are you going to use all of them… The answer is NO, but that's the point. Just like picking the right storage or computing solution, you choose the offering that works for you:
If you want a No-Code style solution to get up and running, Give Amazon Q Business a spin
Want to integrate it with an existing application or workflow… Take a look at Bedrock
Want to do some fine-tuning and customization of a Model, SageMaker (and maybe SageMaker JumpStarts) to the rescue?
Want to start rolling your models from scratch… Get started with Elastic Inference, Inferentia and Deep Learning Images or PyTorch, MxNet or TensorFlow on AWS.
And as always, If you have any questions or queries about any of the AWS AI offerings… or are curious about what AI can do for your organisation… Feel free to contact the Mantel Group team; we'll be more than happy to help. Also, keep an eye on our Blog as we release future articles where we'll dive deeper into specific services and look at how you can get started with them.