Get your Data GenAI Ready! Let's Build a Startup S2E11

In the latest episode of AWS’s "Let's Build a Startup" on Twitch, host Salma Virk, a startup solutions architect at AWS, along with co-host Sinan, Solutions Architect Manager at AWS, delved 😉 into the critical role that data strategy plays when architecting Generative AI (GenAI) powered applications and business models. The session also featured a guest appearance by Sameer Goyal, CEO and Founder of Kidzovo, a pioneering educational platform for children that integrates GenAI to offer interactive and engaging learning experiences.

Sinan opened the discussion by emphasizing that GenAI models are not built out of thin air—they require vast amounts of data sourced from various channels. He outlined three primary sources of data: open and public datasets, private datasets, and data generated internally by businesses. Public datasets, while available to everyone, often require curation and cleaning, which is why some companies opt to purchase these datasets. Private datasets, on the other hand, are often bought to gain deeper insights, such as detailed sports statistics or proprietary business data.

The importance of maintaining and updating these data sources was underscored by Sinan, who highlighted that as time progresses, the relevance of data can diminish. Thus, companies need to consider retraining or fine-tuning their models to keep them updated with the latest information. This process, though costly, is essential for maintaining the accuracy and effectiveness of GenAI applications.

The next part of the discussion focused on where and how to store the vast amounts of data required for GenAI. Sinan recommended starting with simple, scalable solutions such as AWS S3 for long-term storage, and utilizing more specialized storage options like Amazon EFS or FSx Lustre for ephemeral storage. These options ensure that data is accessible, secure, and can be shared across various instances and containers, which is crucial for distributed computing environments.

For long-term storage, Sinan emphasized the importance of data lakes, which allow companies to store structured and unstructured data, making it easier to process and utilize. He also touched on the necessity of data governance and processing, which are integral to maintaining data integrity and usability.

One of the most critical steps in data preparation for GenAI is data cleaning. Sinan explained that data cleaning involves removing corrupt, inconsistent, or duplicate data to ensure that only high-quality data is used for model training. Clean data is essential for accurate and reliable AI models.

Sinan also shared insights into the role of knowledge bases in GenAI applications, drawing on his experience with Alexa. Knowledge bases store factual information and other data types that GenAI models can reference in real-time to provide accurate answers. He highlighted the use of databases like MySQL and DynamoDB, and the growing importance of vector databases, which are particularly useful for familiarity searches in AI applications.

The episode concluded with an interview with Sameer Goyal, who shared the inspiration behind Kidzovo, a cutting-edge app designed to deliver a safe and engaging educational experience for young children. Sameer discussed how Kidzovo leverages AI to create interactive content, ensuring that kids are not merely passive consumers but active participants in their learning journey.

Kidzovo's backend is built entirely on AWS serverless technologies, which allows it to scale efficiently as the platform grows. With over 40,000 registered users across 180 countries, Kidzovo is a testament to the power of a well-executed data strategy combined with the scalability of cloud infrastructure.

If you're eager to explore the innovative world of Kidzovo, you can try the Kidzovo App here, visit the Kidzovo website, or watch more on Kidzovo's YouTube channel. For those interested in deepening their understanding of data strategies, check out AWS resources on data lakes and vector databases. Additionally, there's a valuable blog on designing hybrid AI/ML data access strategies with SageMaker. Don't forget to check out our upcoming episodes and rewatch past episodes to stay updated with the latest insights!

Peak Viewers: 45
Average viewers: 34
Unique Chatters: 12
Messages: 50
CTA Clicks: 32

Security Awareness Disclaimer: Opinions presented by our guests are their own and may not reflect the opinions of AWS. For any considerations of adopting these services and architectural patterns in production, it is imperative to consult with your company specific security policies and requirements. Each production environment demands a uniquely tailored security assessment that comprehensively addresses its particular risks and regulatory standards. If in doubt, reach out to your AWS Account team.