The Art of the Possible: Generative AI
Two thousand fifty-nine.
That's the approximate number of species encompassing the 7.2 billion birds native to North America.
For companies like FeatherSnap, which makes solar-powered bird feeders that stream images and video through an app, it's also an immense data and analytics challenge.
Enter generative AI—a category of artificial intelligence that can ingest and analyze visual information at scale, produce new text and video content in real time, and make complex identifications instantly. Through Amazon Bedrock, startups like FeatherSnap can reinvent customer experiences by transforming resource-intensive tasks (like the identification of thousands of bird species) into an artful solution. Amazon Bedrock offers companies of any size a simple, user-friendly way to build and scale generative AI applications with powerful foundation models (FMs).
AI in the Wild: Bringing Humans and Nature Closer Together
FeatherSnap's mission is to bridge the gap between humans and the natural world using cutting-edge technology, says Kelly Hover, Chief Experience and Marketing Officer at FeatherSnap. "I think there's a sky full of possibilities using generative AI," says Lindsay Bowers, Director of Product Ownership. "By bringing technology and nature together, we're able to introduce people to a world they didn't even know existed."
After installing the FeatherSnap bird feeder that has a built-in camera, users can observe video and still images of birds feeding at the FeatherSnap device on their smartphones. They can then learn more about each identified species and share images with friends and family on social channels. The company's goal, according to FeatherSnap, is to make smartphone tools that draw people toward, rather than away from, nature.
The immediacy of a connection to nature through a mobile device can be a powerful way to connect with nature, Hover believes.
The FeatherSnap team recounts a story of a family whose grandmother was hospitalized and undergoing cancer treatments. The feeder, installed in her backyard, captured images of the birds visiting her garden and automatically sent videos and images to her phone, giving her moments of pure delight when she was unable to view her feathered friends' daily antics in person.
That's why the immediacy and precision of bird identification were critical to the brand's success with its clients and its mission, Hover says. But at launch, the company faced a dilemma.
Investigating AI's potential
When the company started this project over a year ago, they knew that AI species identification was a feature that they did not want to launch without, said Bowers. As birds landed on the bird feeder, they wanted users to be able to choose to identify them on their own or hit an 'I need help' button to have AI do it for them. The challenge, according to Bowers, was determining how sophisticated the AI needed to be.
According to Hover and Bowers, the company had to ensure that the way FMs were trained to interpret visual information could scale easily without sacrificing accuracy. FeatherSnap chose to reach out to Amazon Web Services (AWS). The AWS team knew that the increased use of AI technology tools to track, identify, and evaluate avian behaviors had become important for researchers studying the impact of the environment on wildlife populations. FeatherSnap wanted to leverage that same level of accuracy and data ingestion capacity to launch their app, says Hover.
Gaining Precision and Performance in 45 Days
"It would have been really easy for AWS to look at us and say, 'That's a really cute, fun little project. And we hope that when you get bigger, you come back to us,'" says Hover. "But they didn't; they jumped in with two feet and sat alongside us, literally sat alongside us, and were solving problems with our tech team."
The solution that emerged was a bespoke technical integration—a process that involved FeatherSnap working with a team of experts at AWS. The FeatherSnap team turned to Amazon Bedrock, a fully managed service that helps organizations easily build and scale applications using FMs and generative AI tools.
The team only had 45 days before their launch when they approached AWS, a process that the AWS team acknowledges was "intense." By integrating Anthropic's Claude 3 Haiku—an advanced FM with powerful computer vision capabilities—into their own tech stack through Amazon Bedrock, FeatherSnap was able to enhance its proprietary algorithm's performance and accuracy. FeatherSnap's image recognition system can now distinguish between subtle variations in bird species that might be missed by less sophisticated algorithms.
"When shared in aggregate by users across social media, these images and videos could provide scientists with useful information on bird migratory patterns, according to FeatherSnap."
Leveraging AI for Complex Image Analysis
FeatherSnap's analysis process can also evaluate the context of bird behavior in natural or other habitats and use that data to continually improve the accuracy of their system's algorithms.
"Having an FM like Claude 3 Haiku that can come in and identify birds not just based on color, but also looking at size, beak structure, and plumage is really powerful," says Hover. "If people want to use the AI feature on their cameras, we needed to be able to process those (requests) very quickly and accurately." For example, in one instance, the AWS team notes that the model identified a robin as a chicken during training. The AWS team was able to quickly update model inputs and fix the issue so that trust would not be lost on the customer side.
FeatherSnap's current app identifies birds on sight and sends that information to app users' phones instantly. When shared in aggregate by users across social media, these images and videos could provide scientists with useful information on bird migratory patterns, according to FeatherSnap. Those insights can be valuable to researchers' efforts to track wildlife migration and the carrying capacity of habitats—the maximum number of birds an environment can support without degradation.
How FeatherSnap Used Bedrock and Claude to Build Smarter AI Tools
Using Bedrock, the company was able to privately customize its AI implementation with different data sets like text and images, as well as fine-tune and build agents—or semi-autonomous programs. Those agents, fueled by new data and operating in real-time, can make API calls, query knowledge bases, and execute a range of tasks automatically and on-demand for users.
By leveraging Claude 3 Haiku's generous context window that enables it to process and evaluate large datasets, FeatherSnap can also update new visual image processing capacity on demand. This allows the company to enhance the precision of its own image recognition algorithms easily at scale, says Bowers. "We want to constantly be coming out with new features, and I really think Amazon Bedrock is helping us push those boundaries to make it relevant," says Hover. "This is what makes products timeless in my mind."
"One of the things that AWS has focused on is the democratization of access to generative AI."
Taking Flight: Scaling Innovation
Amazon Bedrock played a pivotal role in the scaling of FeatherSnap's infrastructure, providing a unified API and an intuitive console for managing and deploying powerful FMs for implementing generative AI. This allowed FeatherSnap to focus on innovation rather than the minutiae of infrastructure management, says Hover. "One of the things that AWS has focused on is the democratization of access to generative AI," says Dougherty. "We want users to be able to use tools that are at the right level for the problem they're solving."
production by
Presented by
AWS has unlocked a universe of new options for businesses looking to integrate the power of generative AI apps into their business models.
From intuitive customer experiences to adaptive analytics, Amazon Bedrock is the on-ramp for innovation. Bedrock is a fully-managed service that offers a choice of top-tier foundation models (FMs) from leading AI companies through a single API, along with a broad set of features to build scalable, generative AI applications securely.
Learn more about generative AI and Amazon Bedrock with the upcoming webinar series, Generative AI Unleashed - Transform Your Business.
