Field Notes Powers AI-Driven Research with GPU-Optimized ECS on AWS

Overview

Field Notes, a pioneering technology company, provides a powerful platform for qualitative research, enabling researchers to gather and analyze rich, contextual insights from participants worldwide through self-shot video and multimedia. To enhance the platform’s AI-driven capabilities, particularly the Whisper-based transcription process, Field Notes needed to overcome the challenge of lengthy processing times. Recognizing the scalability and modernization benefits of AWS, they partnered with Lambert Labs to implement a GPU (Graphics Processing Unit)-powered solution on Amazon Elastic Container Service. This transition from CPU (Central Processing Unit)-based to GPU-accelerated workloads significantly reduced transcription processing times, enabling near real-time insights and enhancing the overall research experience for Field Notes’ users.

Opportunity / Customer Challenge

At the heart of Field Notes’ platform lies the ability to provide researchers with AI-powered tools for summarization, time-coded transcription, and tagging, enabling faster and deeper analysis of captured data.   

However, the computational demands of these AI-driven features, particularly the Whisper-based transcription process, presented a significant challenge. While Field Notes recognized the scalability and modernization benefits of deploying their containerized applications on ECS, the default CPU-based processing proved to be a critical bottleneck. The lengthy processing times for video and audio transcriptions were hindering the platform’s ability to deliver real-time or near-real-time insights, impacting researcher workflows and overall user experience.

Field Notes sought to leverage the power of GPUs to accelerate these computationally intensive AI workloads. The opportunity was to transition their Whisper transcription and related AI processes from CPU-based to GPU-accelerated environments within ECS. This transition aimed to dramatically reduce processing times, enabling faster data analysis and empowering researchers to extract insights from their data with unprecedented speed and efficiency. The challenge was to implement this transition seamlessly, ensuring the scalability and reliability of their platform while optimizing costs and maintaining a smooth user experience.

Solution

To overcome the challenge of lengthy processing times for their AI-driven transcription and analysis, Field Notes implemented a high-performance solution on AWS that leverages the power of GPUs. Recognizing the need for scalable and efficient container orchestration, Field Notes chose ECS to manage their application workloads.

Instead of relying on standard CPUs, which proved to be a bottleneck for their computationally intensive AI tasks, Field Notes opted for GPU-accelerated Amazon EC2 instances. Specifically, they utilized g4dn.xlarge instances, powered by NVIDIA T4 Tensor Core GPUs, to significantly speed up the processing of Whisper-based audio and video transcriptions. These GPUs are designed to handle the complex calculations involved in artificial intelligence workloads, resulting in a dramatic reduction in processing time.   

To ensure the solution’s scalability and reliability, Field Notes deployed these GPU-enabled EC2 instances within an AWS Auto Scaling group. This allows the platform to automatically adjust its processing capacity based on demand, ensuring that transcriptions are processed quickly even during peak usage. The Docker container images used to run Whisper and Sidekiq, the application’s background processing framework, are stored in Amazon Elastic Container Registry, providing a secure and scalable repository for their container images.

Security was also a key consideration. Field Notes utilized AWS Secrets Manager to securely store and manage their application credentials, ensuring that sensitive information remained protected.

By leveraging AWS’s GPU capabilities, we’ve been able to use the highest quality of AI-generated transcription (which wouldn’t have been possible with CPU), enabling us to deliver faster insights to our researchers. Lambert Labs’ expertise was instrumental in guiding us through this transition and ensuring a smooth implementation. (Tim Diggins, CTO / Co-founder, Field Notes)

Outcome

By implementing a GPU-powered solution on ECS, Field Notes successfully transformed its AI-driven research platform, delivering significant improvements in processing speed and overall user experience. The transition from CPU-based to GPU-accelerated workloads resulted in up to a 95% reduction in video and audio transcription processing time, dramatically accelerating the analysis workflow for researchers.

This enhanced speed has enabled Field Notes to provide near real-time insights from captured video and audio data. Researchers can now access key findings from their research much faster, significantly reducing the time required to extract valuable insights. This acceleration has not only improved researcher satisfaction but also streamlined the entire research cycle, allowing for quicker iteration and more agile research projects.

The ability to process transcriptions and generate AI-driven summaries and tags at a significantly faster pace has also opened up new possibilities for Field Notes’ business. In a market where rapid insights are crucial, Field Notes has gained a competitive advantage by delivering faster, more efficient research capabilities. This enhanced efficiency has enabled Field Notes to handle larger volumes of research data, supporting their growth and expansion into new markets.

Furthermore, the scalability and reliability of the AWS-based solution, powered by GPU-enabled EC2 instances and Auto Scaling, have ensured that Field Notes can consistently deliver high-performance analysis, even during periods of peak demand. This reliability has strengthened Field Notes’ position as a leading provider of AI-driven qualitative research tools, allowing them to focus on innovation and growth rather than infrastructure management.

Using GPUs together with containerization is a powerful combination for AI workloads. We were able to deliver real value to the customer via AWS’s offerings in these areas. (George Lambert, Founder & CEO, Lambert Labs)

About Field Notes

Field Notes is a technology company that provides a cutting-edge platform designed to revolutionize qualitative research. By leveraging mobile technology and AI-powered tools, Field Notes enables researchers to capture and analyze rich, contextual insights from participants worldwide. The platform simplifies the process of gathering multimedia data, including self-shot video, audio, and text, from individuals in diverse locations and environments. Field Notes empowers researchers across various sectors to conduct in-depth studies, gain a deeper understanding of human behavior, and accelerate the pace of discovery.