FeatherSnap Offers AI-Powered Bird Watching Using Amazon Bedrock and Anthropic’s Claude

FeatherSnap offers a smart bird feeder that gives customers a window into the wild. With the accompanying mobile app, users can identify local bird species that visit the feeder, learn fun facts about them, and save their photos into digital bird books.

FeatherSnap already had a dataset and workflow to support its manual bird-identification process, which required users to enter a bird’s color, size, and other details to receive results. However, before launching its first product, the FeatherSnap Scout bird feeder and app, the company wanted to power its identification feature with generative AI to enhance the user experience for birders.

During product development, FeatherSnap began building its own generative AI identification model from scratch, but it didn’t have enough core data to accurately identify the full scope of the nondomestic, North American bird species. To improve bird classification accuracy, FeatherSnap decided to use a large language model (LLM) and started writing prompts using a third-party provider. Seeking superior accuracy and price performance, the company quickly pivoted to using Anthropic’s Claude 3 Haiku in Amazon Bedrock. Claude 3 Haiku is an exceptionally fast and compact model with vision and text capabilities, and it can generate near-instant responses. “All our core services are built on AWS architecture,” says Michael Leo, director of delivery at FeatherSnap. “When we learned about the capabilities of Amazon Bedrock, it was a simple decision to use the service because we already had a deep understanding of and trust in AWS services.”

To meet the approaching product release date, FeatherSnap collaborated closely with experts from AWS to validate the generative AI feature and rapidly iterate. FeatherSnap’s team had limited experience using LLMs, which can have a steep learning curve initially. Working alongside AWS, the team gained speed and optimized its prompts to return more accurate results to customers. “We were able to expedite our efforts to meet consumer expectations,” says Leo. “We had pretty robust expectations, so we likely wouldn’t have achieved resolutions as quickly as we did had it not been for AWS.”

In June 2024, FeatherSnap released its smart bird feeder and app less than 90 days after its proof of concept for the generative AI bird-identification feature. “We got up and running quickly with a reliable product to deliver to our consumers,” says Lindsay Bowers, director of product ownership at FeatherSnap. “We know that every time someone clicks the AI button, we will deliver a good consumer experience, and this gives us peace of mind.”

When a bird lands on the feeder, the integrated camera takes photos and sends them to the FeatherSnap app for the user to view. If the user chooses to identify the bird using the generative AI feature, the image is sent to Amazon Bedrock with a prompt to identify the bird, taking into account the user’s location and other details. The images are stored and accessed using Amazon Simple Storage Service (Amazon S3), an object storage service built to retrieve virtually any amount of data from anywhere. “Using Amazon S3, we can store customer data in one place without sending photos to different places and worrying about security across multiple services,” says Bowers.

To identify bird species correctly, FeatherSnap first worked with an ornithologist to compile a robust library of photographs of identified birds. Then, the company tested and refined its prompting strategy by validating that the results from Claude 3 Haiku in Amazon Bedrock matched expectations. Throughout the process, FeatherSnap worked closely with experts from AWS to troubleshoot and optimize results. By the product’s release date, FeatherSnap achieved 89 percent accuracy in identifying nondomestic birds in its supported region of North America. Using Claude 3 Haiku in Amazon Bedrock, FeatherSnap also optimized price performance. “Claude 3 Haiku was the right solution available at the right time and place for serving the needs of our organization and presenting what we wanted to our consumers,” says Leo.

FeatherSnap plans to continue enhancing its app with new features that help users enjoy nature and gain information about caring for the birds around them. For example, FeatherSnap is currently developing more gamification for the app to further engage users. The company is also considering expansion into additional regions, such as Australia and Europe. “We’ve proved that the AI identification that is built on Amazon Bedrock works, which opens doors to expand into different areas,” says Bowers.