Enhancing Fan Engagement with Generative AI Commentary for Professional Golf

Objective

The PGA TOUR partnered with NLP Logix and AWS to design and deploy an enterprise-grade generative AI system capable of delivering real-time, contextual commentary for every shot during professional golf tournaments

With tournaments featuring up to 156 players and nearly 30,000 shots per event, the objective was to extend fan engagement beyond traditional broadcast coverage without proportionally increasing traditional commentary teams.

Challenge

Professional golf presents a uniquely complex coverage environment:

  • Multi-day tournaments across expansive courses
  • Thousands of simultaneous shot-level moments
  • Nearly 256,000+ ShotLink datapoints captured weekly
  • Over 100,000 weekly TOURCAST users expecting real-time updates

Traditional broadcast models cannot narrate every shot in real time. The PGA TOUR needed a scalable, low-latency solution that could transform structured event data into high-quality, contextual narratives while maintaining editorial standards.

Solution

A cloud-native, event-driven generative AI commentary pipeline was created and integrated directly into TOURCAST this included:

Data-Grounded Generative AI Architecture

  • Deterministic data in, narrative out: ShotLink shot events serve as the factual backbone for commentary generation.
  • Two-stage prompting: First generate fact-based descriptions, then enrich with tournament context and player history to explain why the shot matters.
  • Production orchestration layer: Standardized model invocation, prompt management, logging, and portability across foundation models.

Real-Time Event Processing

Shot events flow through an AWS messaging backbone:

  • Amazon SNS → Amazon SQS → AWS Lambda
  • API Gateway → Amazon ECS service
  • Amazon Bedrock for LLM invocation

Context & State Management

  • Amazon DynamoDB stores shot history, leaderboard history, and evaluation artifacts (“truth deck”)
  • Commentary is published back to SNS for downstream TOURCAST integration
  • Amazon CloudWatch dashboards monitor system health and GenAI-specific metrics

Automated Quality Validation at Scale

To uphold editorial standards across ~32,000 shots per event:

  • Fact extraction and source comparison against ShotLink
  • Cosine similarity scoring against gold-standard examples
  • LLM-as-a-Judge grading for accuracy, hallucination risk, and toxicity
  • Combined validation thresholds before publication
golfer putting

Results

  • Enabled real-time, contextual shot-level commentary inside TOURCAST
  • Scaled to support tournaments with 30,000+ shots per event
  • Successfully transformed 256,000+ weekly ShotLink datapoints into narrative content
  • Maintained high-quality standards through automated validation
computer chip

Tech Stack

Generative AI & Orchestration

  • Amazon Bedrock
  • Anthropic Claude 3.5 Sonnet and Claude 3.5 Haiku
  • LLMKit orchestration framework

Event-Driven Infrastructure

  • Amazon SNS
  • Amazon SQS
  • AWS Lambda
  • Amazon API Gateway
  • Amazon ECS

Data & State Management

  • Amazon DynamoDB

Monitoring & Governance

  • Amazon CloudWatch dashboards and alarms
  • CI/CD practices for prompts and evaluation

Data Sources & Integration

  • ShotLink (shot-level event data)
  • TOURCAST (fan-facing application)
View More Success Stories