Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Agentic generative AI for media content discovery at the national football league

Henry Wang, Md Sirajus Salekin, Jake Lee, Ross Claytor, Shinan Zhang, Michael Chi

TL;DR

The paper addresses the challenge of finding relevant NFL game footage in vast Next Gen Stats archives by introducing an agentic generative-AI workflow that translates natural-language queries into structured API calls. It leverages text-to-SQL and Retrieval-Augmented Generation, semantic caching, and schema routing within an AWS-based pipeline (Bedrock, LangGraph) integrated with the NFL's MAM for asset retrieval. Results show over 95% accuracy on 100 QA prompts, onboarding of ~250 users in the first month, and dramatic reductions in search time from ~10 minutes to ~30 seconds, with significant improvements for both simple and complex queries. This work enhances operational efficiency for media teams and lays groundwork for fan-facing discovery tools, accelerating content creation and storytelling.

Abstract

Generative AI has unlocked new possibilities in content discovery and management. Through collaboration with the National Football League (NFL), we demonstrate how a generative-AI based workflow enables media researchers and analysts to query relevant historical plays using natural language rather than traditional filter-and-click interfaces. The agentic workflow takes a user query as input, breaks it into elements, and translates them into the underlying database query language. Accuracy and latency are further improved through carefully designed semantic caching. The solution achieves over 95 percent accuracy and reduces the average time to find relevant videos from 10 minutes to 30 seconds, significantly increasing the NFL's operational efficiency and allowing users to focus on producing creative content and engaging storylines.

Agentic generative AI for media content discovery at the national football league

TL;DR

The paper addresses the challenge of finding relevant NFL game footage in vast Next Gen Stats archives by introducing an agentic generative-AI workflow that translates natural-language queries into structured API calls. It leverages text-to-SQL and Retrieval-Augmented Generation, semantic caching, and schema routing within an AWS-based pipeline (Bedrock, LangGraph) integrated with the NFL's MAM for asset retrieval. Results show over 95% accuracy on 100 QA prompts, onboarding of ~250 users in the first month, and dramatic reductions in search time from ~10 minutes to ~30 seconds, with significant improvements for both simple and complex queries. This work enhances operational efficiency for media teams and lays groundwork for fan-facing discovery tools, accelerating content creation and storytelling.

Abstract

Generative AI has unlocked new possibilities in content discovery and management. Through collaboration with the National Football League (NFL), we demonstrate how a generative-AI based workflow enables media researchers and analysts to query relevant historical plays using natural language rather than traditional filter-and-click interfaces. The agentic workflow takes a user query as input, breaks it into elements, and translates them into the underlying database query language. Accuracy and latency are further improved through carefully designed semantic caching. The solution achieves over 95 percent accuracy and reduces the average time to find relevant videos from 10 minutes to 30 seconds, significantly increasing the NFL's operational efficiency and allowing users to focus on producing creative content and engaging storylines.

Paper Structure

This paper contains 15 sections, 7 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Data
  4. Schema
  5. Question and Answer Pairs
  6. Methodology
  7. Agentic Workflow
  8. System Infrastructure
  9. Results
  10. Conclusion
  11. Data
  12. Schema
  13. Question and Answer Pairs
  14. System Interface
  15. System Architecture

Figures (7)

  • Figure 1: Overall diagram of the Agentic workflow
  • Figure 2: Links to retrieved assets embedded in React frontend
  • Figure 3: The solution clarifies with the user when their initial prompt contains a player name that is tied to multiple distinct players in the database.
  • Figure 4: Reasoning step and formulated API.
  • Figure 5: Natural language response of the query.
  • ...and 2 more figures