Amazon Bedrock Builders’ Challenge

Generative artificial intelligence (AI) is already reinventing customer experiences and unlocking a wide range of use cases. But here at Amazon Web Services (AWS), we know the most exciting possibilities are yet to come. That’s why we invited startups to test the limits of what generative AI can do—introducing the Amazon Bedrock Builders’ Challenge.

Startups that could demonstrate advanced technical implementation, adoption of AWS services, innovation, and potential customer impact would be rewarded with US$100,000 in AWS credits.

Alongside US$5,000 in AWS credits for experimentation and iteration, contestants were given just one rule: their production workloads had to be powered by AWS generative AI primitives—Amazon Bedrock, Amazon SageMaker, and Amazon EC2 Capacity Blocks. We’ve seen some truly standout entries—from speeding up large language model (LLM) inferencing to redefining how plastic is recycled.

Today, we’ll introduce you to our five winners and show how their creativity and ingenuity earned them a top spot.

Circleback helps teams get the most out of every meeting

Circleback helps teams get the most out of every meeting–whether it's online or in-person–with AI-powered notes, action items, automations, and powerful insights. Circleback was founded in 2023 by Ali Haghani and Kevin Jacyna—University of British Columbia alumni with previous collaborations under their belt. Previously, Ali led the first integration of large language models (LLMs) into Stripe’s support product, while Kevin worked on predictive analytics at Tableau.

Harnessing their combined experience together with Amazon Bedrock, Circleback developed an extensible custom insight identification pipeline as its competition entry. This allows users to extract structured data from their conversations at scale in over 100 languages. Today, the startup—along with its Amazon Bedrock-powered LLM—has secured TechCrunch recognition and was TIME Magazine's top pick for AI note-taking.

How Clearly AI automates privacy and security reviews

Started by former Amazon engineers Emily Choi-Greene and Joe Choi-Greene in 2024, Clearly AI helps businesses ensure compliance with industry standards and internal policies, so they can focus on what matters most. The startup empowers customers to reduce review backlog and cycle time by 90 percent, while providing clear risk visibility and actionable recommendations for leadership and developers respectively.

For its entry, Clearly AI migrated from OpenAI GPT 4.0 to Claude 3.5 Sonnet and built a custom orchestration layer to help break down complicated assessments into logical stages, while maintaining context throughout the process. Migrating to Claude 3.5 Sonnet also allows Clearly AI to support its enterprise customers’ bring-your-own-cloud (BYOC) requirements with the use of Amazon Bedrock.

Most recently, the startup has expanded to help customers within regulated industries, for use cases such as threat modeling, vendor risk reviews, and privacy impact assessments. With Clearly AI, a leading EV manufacturer was able to quadruple their review capacity while maintaining the same high quality.

Rubbrband’s one-stop-shop for media generation

For AI image generation tools, it can be challenging to create high-quality assets that consistently take into account specific user preferences and style criteria. Founded in 2022 by former UC Berkeley roommates Jeremy Lee, Darren Hsu, and Abhinav Gopal, Rubbrband is on a mission to empower creatives—from beginners to professional production studios—with increased personalization.

During the competition, Rubbrband used prompt engineering with Amazon Bedrock to implement and quickly iterate on its AI artist, Roi. By focusing on user-controlled workflows, Roi is able to consistently generate images tailored to individual preferences.

Leveraging Amazon Bedrock on Amazon EC2, Roi’s technical implementation is well underway, with image creation and editing features already fully operational. Today, Rubbrband has doubled its growth rate and surpassed 50,000 total users.

Typewise’s AI assistant powers customer service and sales excellence

A remote-first startup with offices in the US and Switzerland, Typewise aims to automate and optimize enterprise customer service. Combining customizable AI agents with an intelligent assistant for human agents, Typewise enables rapid automation gains while maintaining a personal touch in customer interactions. Founded by David Eberle and Janis Berneker in 2020, Typewise received backing from the prestigious Y Combinator accelerator program.

For its competition entry, Typewise used Amazon SageMaker and Low-Rank Adaptation (LoRA) to train its LLMs, with test deployments held on SageMaker endpoints. It also experimented with running multiple parallel pipelines to optimize model latency at scale. As a result, Typewise discovered multiple ways to speed up model inference, such as speculative decoding and key-value (KV) caching.

Today, Typewise has moved beyond initial product validation and secured over 40 enterprise contracts in Europe and the US. The startup has consistently achieved an 80 percent pilot-to-annual license conversion rate with zero churn.

Birch Biosciences: Reinventing the plastic recycling process

With 400 million tons of plastic produced every year—and only 9 percent of it recycled—Birch Biosciences focuses on making plastic recycling more economical, sustainable, and circular. Founded in 2022 by Johan Kers and Emily Duncan, the startup engineers enzymes that break down plastics into chemical building blocks, which can then be remanufactured into high-quality products made from 100 percent recycled content.

As part of Amazon Bedrock Builders’ Challenge, Birch Biosciences developed an enzyme engineering platform that leverages design-build-test-learn (DBTL) cycles. This platform uses large volumes of microfluidics, high-performance liquid chromatography, and next-generation DNA sequencing (NGS) data to train and refine enzyme design models.

In 2024, Birch Biosciences met key technical milestones for enzyme performance in a plastic recycling process. This demonstrated the startup’s AI innovation flywheel and achieved critical metrics for scaling and commercializing its technology. Since then, Birch Biosciences has relocated to a brand-new 6,500 square-foot facility that will house its first recycling pilot plant.

Additionally, the startup has received notable awards from the National Science Foundation, Department of Energy, and Defense Advanced Research Projects Agency to develop its recycling technology, as well as to expand research into different types of plastic.

Discover what’s possible with AWS generative AI

AWS is home to over 280,000 startups—including 80 percent of Y Combinator startups—with a range of programs and resources dedicated to helping them launch and grow successfully. We’ve been amazed by the level of creativity and technical expertise displayed by our Amazon Bedrock Builders’ Challenge winners. 

AWS helps startups bring their ideas to life through AWS Activate. As you build and scale your business, Activate Credits grow with you to support your changing needs. If you’re an AWS Activate member, you may be eligible for up to $100,000 in credits to help offset your AWS bill. Apply for Activate Credits today.