Building a Personalized Sports Concierge Agent with AI
How Chester Ismay automated his sports viewing recommendations using AI
1Automating sports preferences
Chester Ismay created a structured preferences file to identify teams he follows and the context behind his choices. He built a simple web interface for editing preferences instead of manually handling JSON files. The system tracks multiple leagues including WNBA, NFL, NBA, and Premier League. This foundation allows the agent to understand which games might interest him without manual input each week.
2Data integration setup
The agent pulls schedule information from ESPN and other sources stored in GitHub repositories. A Node.js tool combines these schedules with Chester's preferences, creating a comprehensive dataset. This approach prevents the agent from needing to search for each game individually. The system maintains updated information while respecting data source structures and formats.
3Policy-driven decision making
The CONCIERGE.md file defines the agent's core logic and rules for recommendations. It specifies how to handle completed tournaments, duplicate matchups, and output formatting. Chester refines the policy through iteration when outputs miss expectations. The document serves as both specification and debugging reference, with clear instructions for edge cases and special scenarios.
4Secure execution environment
Chester configured Claude Code to operate with strictly limited permissions, accessing only necessary files and tools. This security-conscious approach prevents unintended system access while allowing the agent to perform its designated tasks. The setup demonstrates how to balance functionality with security in personal automation projects.
5Scheduled delivery system
The system runs automatically every Wednesday using Mac's launchd scheduler. Recommendations get delivered to Chester's phone via ntfy push notifications. This hands-off approach ensures consistent delivery without manual intervention. The scheduling accounts for processing time while aligning with Chester's weekly planning routine.
6Quality control measures
Chester implemented verification against source schedules and uses multiple data sources to catch discrepancies. He added explicit time zone handling after UTC defaults caused date misalignments. These checks maintain recommendation accuracy as schedules change. The system evolves through continuous testing and validation against real-world data.
7Iterative improvement process
When recommendations don't meet expectations, Chester revisits the policy file to refine instructions. This feedback loop allows gradual improvement without system overhauls. Adjustments might include clearer goal definitions or better alignment with project objectives. The process demonstrates practical AI tuning for personal use cases.
8Future of personal AI agents
This project showcases how professionals can build customized AI solutions for niche needs. The sports concierge exemplifies practical AI application beyond generic tools. Its success suggests potential for similar agents in other personal domains. The approach balances technical sophistication with maintainability for non-enterprise use cases.
Original article: O'Reilly Radar — Building a Personalized Sports Concierge Agent with AI