AI creator analysis and search

1. Database of TikTok Creators and Posts

• Scale: We compiled an extensive database containing 500 million TikTok creators and nearly 1 billion posts.
• Filtering: From the initial pool of 500 million creators, we filtered down to 7 million who met specific criteria relevant to KYRA's needs. This filtering ensured the data was both manageable and highly targeted.

2. AI-powered post Description Prototype

• Objective: To enable advanced search capabilities, we experimented with a prototype that could describe individual posts in detail. This allowed users to search for specific visual or contextual elements within videos.
• Example Use Case: A user could query "yellow bag," and the system would return all videos where a yellow bag appeared. Similarly, users could ask for creators who wear "smart casual clothes."
• Challenge: While this prototype demonstrated the potential for highly granular searches, running this level of analysis across 1 billion posts proved too expensive. Due to the high costs of scaling this experiment across the entire dataset, it was not implemented in production.

3. AI-Driven Search Query Optimization

• Solution: Instead of relying on detailed post descriptions, we developed an AI system capable of generating optimized search queries using logical operators (e.g., ORs and ANDs).
• How It Works: The AI writes complex queries that efficiently search through the database based on user input. This approach balances functionality with cost-effectiveness, allowing KYRA to retrieve relevant results without the need for exhaustive post-level analysis.

Outcome

This combination of filtering, prototyping, and AI-driven search query optimization provided KYRA with a scalable and efficient solution to navigate their vast creator database while maintaining flexibility for specific use cases.