Build and deploy production-quality generative AI apps (AIM257)

Building and Deploying High Quality Generative AI Applications

Introduction

• The focus of DataBricks is on helping customers build high-quality AI applications that can be put into production and critical business use cases.
• Traditional "one-click" AI solutions may not perform at the level required for real-world deployment.
• DataBricks aims to provide a unified platform to manage and govern all aspects of building and deploying AI agents.

About DataBricks

• DataBricks is a data intelligence platform that helps manage and govern data, models, agents, and tools in a unified system.
• DataBricks has been recognized as a leader in data and AI by industry analysts.
• The company works with customers across various industries to build strategic AI capabilities.

The Evolution of Generative AI

• There has been significant progress in the performance of foundation models, both private and open-source, over the past year.
• DataBricks' view is that the best model to use is the one that performs best for the specific use case, rather than pushing a single proprietary model.
• The challenge is that most generative AI projects fail to make it to production, often due to issues with quality, cost, and governance.

Building Data Intelligence Agents

• Successful data intelligence agents need to be able to reason over structured and unstructured data, use custom evaluation metrics, and be governed in terms of data and model access.
• DataBricks provides a unified platform to ingest and transform data, build agents using any model, and deploy and govern these agents.

Deploying and Governing Agents

• DataBricks offers model serving endpoints to bring in external models and govern their usage.
• Agents can be augmented with "tools" - Python functions that can be registered and used by the agents to take actions.
• The platform provides observability and tracing to understand the agent's decision-making process.

Evaluating and Improving Agents

• DataBricks helps customers build custom evaluation datasets and use LLM judges to assess the performance of agents at each step.
• The platform allows subject matter experts to provide feedback and refine the ground truth data used for evaluation.
• Customers can fine-tune models and optimize the agent's performance using the DataBricks platform.

MasterCard's Experience

• MasterCard's R&D team has been experimenting with generative AI and building a production-ready customer service agent.
• The agent is built using a model factory approach, with multiple models for different domains, and a human-in-the-loop system for continuous improvement.
• MasterCard has seen significant improvements in the agent's performance, from the low 60% range to the high 80% range, by leveraging DataBricks' tools and capabilities.

Next Steps

• Visit the DataBricks booth at the event or reach out to your account executive to learn more and schedule a meeting.
• Check out the DataBricks AI Cookbook for more information and try out the DataBricks Playground to experiment with the technology.