Churchill Downs Races Ahead with AI: How Paddock PRO Uses Amazon Bedrock to Help Transform the Horse Player Experience

This blog post was co-authored by Brian Rickabaugh, AI Solution Architect, and Harold Reed, Sr. Director of Architecture at TwinSpires

INTRODUCTION

In the world of horse racing, few events are as prestigious as the Kentucky Derby. Since 1875, the race has been held at historic Churchill Downs in Louisville, Kentucky.

Churchill Downs Technology Initiatives, a Churchill Downs Incorporated (CDI) property, is named after its iconic Louisville racetrack, and has become a leader in digital gaming technology. Its TwinSpires platform supports gaming for hundreds of races daily all around the world. CDI’s infrastructure powers millions of transactions during these marquee events, handling everything from funding and real-time wagering to customer account management, while preserving the tradition and excitement that has made Churchill Downs synonymous with excellence in thoroughbred racing for generations.

THE FUTURE OF WAGERING

The digital gaming landscape has experienced explosive growth in recent years, driven by state-by-state legalization across the United States. According to the American Gaming Association, the legal sports betting industry in the United States reached just shy of $17 billion in revenue in 2025, with significant expansion opportunities still emerging as additional states consider legalization. This growth has created new opportunities for an industry that has operated legally for far longer – horse racing.

What distinguishes horse race wagering from conventional sports betting is its pari-mutuel system, a pool-based betting model where all wagers of a particular type are pooled together, and payouts are calculated based on the total pool after the track takes its commission – rather than betting against a bookmaker at fixed odds. This fundamental difference means that CDI’s corporate interests are aligned with those of its customers: excitement, engagement, and wagering activity drive customer success — rather than customer losses.

In this unique landscape where horse racing operators compete not only with each other but increasingly with the broader sports betting industry, the ability to innovate through advanced analytics, real-time data processing, and emerging technologies like generative artificial intelligence have become more essential for maintaining relevance and driving growth in an industry experiencing significant transformation.

SOLUTION OVERVIEW

For the 2026 Kentucky Derby, TwinSpires introduced Paddock PRO, an agentic AI solution that assists players by delivering continuously updated information related to horses, connections, track conditions, and more. The goal of Paddock PRO is to provide players with the information and analysis they need to make wagers, from surfacing relevant statistics and insights to helping players evaluate potential wager structures. Players retain full control over whether, and how to, place their wagers.

Figure 1: Paddock PRO Architecture

Players access Paddock PRO via the TwinSpires website and are presented with a chat interface through which they are able to get information. Free-tier players have access to curated prompts tailored to a daily featured track and race. Premium players can go beyond curated prompts and ask their own bespoke prompts, getting personalized answers tailored to exactly what they want to know.

ARCHITECTURE

For Paddock PRO, Amazon Elastic Container Service (ECS) hosts the front-end logic for accepting and relaying queries from the player. When a player prompt is submitted, the application passes the player’s prompt to Amazon Bedrock Agents, the primary service powering this interaction. Amazon Bedrock Agents is an agentic solution that manages the planning and orchestration needed to formulate a response to a prompt – from reasoning a solution plan, enriching the context with available information, and submitting the inference to the foundation model, a large language model (LLM). In this case, the Bedrock Agent is utilizing both Amazon Bedrock Knowledge Bases and action groups to augment the player’s prompt with real-time, contextually relevant information.

During each step of the orchestration loop, Amazon Bedrock Agents interprets the player’s input alongside the accumulated conversation context and uses the foundation model’s reasoning capabilities to determine whether the next best action is to invoke an action group, such as calling an API or executing a AWS Lambda function, or to query an associated knowledge base for additional context, dynamically selecting the optimal path based on the task at hand.

The agent can query Amazon Bedrock Knowledge Bases for Retrieval-Augmented Generation (RAG), pulling in structured and unstructured data from documents stored in Amazon S3 as well as content hosted on the TwinSpires website and other integrated proprietary data sets.

When appropriate, the agent may invoke action groups to retrieve dynamic prompt-specific information from external systems and APIs. In this use case, action groups allow the agent to access a range of TwinSpires data sources, including Amazon Athena tables backed by Amazon S3-resident datasets, custom CDI APIs, and third-party APIs such as current weather information.

With information made accessible through action groups and historical data drawn from Amazon Bedrock Knowledge Bases, the Bedrock Agent constructs a response to the player prompt – combining all necessary information into a coherent, optimized conversation sent to the foundation model. Amazon Bedrock’s model catalog helps provide Paddock PRO the ability to readily change the underlying model as newer and more capable models become available, helping ensure the application is always using the right model for the task.

Figure 2: The Paddock PRO interface

Within seconds, Paddock PRO receives player prompts, enriches them with real-time and historical data, and delivers a generated response powered by a state-of-the-art LLM. Whether a player is asking about specific information or needs assistance with crafting a competitive wager, Paddock PRO places the depth of TwinSpires’ industry knowledge directly into the player’s hands.

CONTEXT AND CONVERSATION MANAGEMENT

The horse racing domain is filled with statistics to help any player find their angle. From pedigree, connections, split times, and trouble trips to coat color and weather forecasts, Paddock PRO provides it all.

Paddock PRO’s knowledge sphere is carefully crafted to provide wagering assistance to a diverse population of players – from newcomers to serious horse players and experienced handicappers. Paddock PRO offers more than five hundred individual statistics covering a broad set of dimensions.  To deliver this experience conversationally, Bedrock Agents’ memory and session management maintain conversation history throughout a player session and continuously augments the orchestration prompt with prior interactions to preserve context across multiple turns. This enables Paddock PRO to recall a player’s earlier questions, refine its reasoning based on previous observations, and deliver increasingly personalized insights, helping transform what would otherwise be stateless model invocations into a coherent, multi-turn wagering assistant experience.

GOING FORWARD

As the generative AI landscape continues to evolve at a rapid pace, the opportunities for innovation have never been more exciting. Amazon Bedrock’s capabilities—from its diverse selection of foundation models and seamless API access to powerful features like Amazon Bedrock Agents, Knowledge Bases, and Guardrails—help organizations move from concept into production-ready solutions with remarkable speed and agility.

Looking ahead, Churchill Downs Technology Initiatives is eager to explore Amazon Bedrock AgentCore, which is designed to help unlock even more sophisticated agentic workflows and autonomous capabilities. With Amazon Bedrock’s managed infrastructure, built-in security controls, and continuous delivery of cutting-edge features, the service helps empower teams to focus on creating value for players, making it the ideal foundation for sustained AI-driven transformation in the horse racing industry and beyond.