AI-Powered Olympics Pipeline: Indonesia Ranks Top 5 Globally

The Olympics generates thousands of video assets in a matter of weeks. Each asset required accurate tagging with athlete names, home countries, sports categories, and event numbers to be discoverable on Google. Doing this manually was not feasible at the required scale and speed. The challenge was compounded because AI at the time was less mature than today: edge cases were frequent, and prompts needed constant refinement to produce consistent results.

Designed and built an AI automation system using GPT-4 mini as the detection engine. The model was connected to Google Sheets via Google Apps Script, letting the team work within a familiar interface without adopting new tooling. Each content row was processed automatically: the AI detected athlete names, home countries, sports categories, and competition numbers from titles and metadata. The system was built iteratively, with extensive prompt engineering to handle variations in international athlete names and inconsistent content formats.

2,834 assets were successfully uploaded and tagged to Google's systems (ViTA and Olympics Onebox). Indonesia ranked in the Top 5 countries worldwide by Olympics content contribution. Google formally recognized Vidio as one of the top global broadcasters by content volume. This achievement received direct acknowledgment from the Google team, passed on to the content team and Vidio's leadership.