AWS re:Invent 2025 -NFL Fantasy AI: Zero to Production in Weeks w/ Bedrock and Strands Agents-SPF304

Building a Production-Ready Fantasy AI Assistant with AWS and the NFL

Overview

• The NFL and AWS collaborated to build a production-ready fantasy AI assistant that could provide expert-level analysis and recommendations to fantasy football managers in just 8 weeks.
• The key goals were to deliver accurate, fast, and secure fantasy advice powered by exclusive NFL data and AI capabilities.
• This presentation covers the architectural decisions, technical challenges, and lessons learned in taking this AI assistant from zero to production.

Architectural Decisions

Agentic Architecture

• The team chose to build an agentic system using the Strands Agents framework, which handles session management, prompt management, and integration with multiple language models.
• This allowed the agent to reason, plan, and take actions autonomously to answer complex fantasy-related questions.

Model Context Protocol (MCP)

• MCP was used as the semantic data layer, separating the agent logic from the data sources.
• This allowed the agent and data layer to be scaled independently and enabled reuse of the data layer for future agents.

AI-Assisted Coding

• Due to the tight timeline, the team leveraged AI coding assistants to speed up learning new frameworks, fill knowledge gaps, and automate undifferentiated code like test suites.
• This allowed the team to focus on the core agent logic and architectural decisions rather than spending time on boilerplate code.

Technical Challenges and Lessons Learned

Building the "Agentic Playbook"

• Challenges in understanding the complex and contextual NFL NextGen Stats data.
• Solution: Created a semantic stats dictionary using LLM-assisted refinement to provide the agent with just the right data at the right time.

Consolidating Tools

• Initial approach of creating a tool for each use case led to fragmented and inefficient agent behavior.
• Solution: Consolidated tools based on data boundaries, allowing the agent to request richer responses in fewer calls.

Handling Production Resilience

• Concerns about throttling and service capacity issues on game day.
• Solution: Implemented a fallback provider using a secondary language model to intercept and handle throttling, ensuring a consistent user experience.

Predicting Emergent Behavior

• Challenges in understanding how the agent would behave in production with real-world user inputs.
• Solution: Extended the Strands Agents framework to provide per-turn reasoning insights, allowing the team to observe and refine the agent's decision-making patterns.

Caching for Performance

• The vast and token-rich nature of the NFL data required careful caching strategies.
• Solution: Implemented a simple sliding window cache for the most heavily used MCP tool calls, resulting in a 2x increase in throughput.

Business Impact and Next Steps

• The fantasy AI assistant was successfully launched on the NFL Pro platform, providing expert-level analysis and recommendations to fantasy football managers.
• Key metrics achieved:
  • 90% analyst-approved accuracy
  • Sub-5 second initial response times
  • Sub-30 second complex analysis responses
• The NFL plans to leverage the fantasy AI assistant to boost the productivity of their internal analyst team, using the AI to help generate weekly insights and analysis.
• Future plans include integrating the assistant with user preferences, league data, and feedback loops to further personalize the experience.

Conclusion

• The NFL and AWS were able to build a production-ready fantasy AI assistant in just 8 weeks by prioritizing practical, pragmatic decisions over perfection.
• Key lessons include:
  • Focus on building intelligence first, then worry about delivery
  • Leverage open-source frameworks and AI-assisted coding to accelerate development
  • Expect and plan for production challenges, don't try to predict the unknown
  • Implement simple, effective patterns for caching and resilience
• The success of this project demonstrates the power of agentic AI systems to transform user experiences, even in complex, data-rich domains like fantasy sports.