View all blogs

Announcing SEEKER-1: state-of-the-art model for querying structured data with natural language

Seek AI is excited to announce the public availability of SEEKER-1, our production natural-language-to-SQL engine that powers Seek AI. The model uses several methods to accurately transform text to SQL.

MiniSeek, a simplified version of SEEKER-1, is the first model to surpass 90% accuracy on the Spider leaderboard and top the leaderboard for exact matching evaluation. SEEKER-1 was found to achieve even higher accuracy than MiniSeek when evaluated on the Spider dev dataset.

With an API and flexible infrastructure, SEEKER-1 can be integrated into software to power generative AI for data. Also available with the Seek AI web app or Slack integration.

SEEKER-1 Model Specs:

  • A system of models for natural language processing of structured data
  • Unique “semantic network” architecture to support reinforcement and in-context learning
  • Proprietary large language models
  • Option to integrate with 3rd-party large language models in lieu of private models
  • API to power integration into other software

View all blogs