Ep 19: Tome CEO Keith Peiris on Disrupting Powerpoint and Scaling To Millions of Users

Model Architecture

Tome's AI model architecture is designed for flexibility and scalability. Keith Peiris explains that they rebuilt Tome's architecture to allow for hot-swapping models within a day, ensuring adaptability to rapid changes in AI technology. This approach includes capacity management to prioritize pro users during peak times and a claim classifier to combat hallucinations by verifying claims through a web crawler 1. Keith also discusses the decision to use off-the-shelf models initially, with plans to build proprietary models as they gather more data, enhancing the user experience with faster and more accurate outputs 2.

We actually rebuilt Tome's architecture with models so we can hot swap a model within a day.

--- Keith Peiris

This strategy allows Tome to efficiently manage resources and improve the quality of generated content.

   

Model Evaluation

Evaluating AI models at Tome involves both internal and external processes. Keith emphasizes the importance of internal evaluation, where the team ensures they like the output before presenting it to users. They focus on conciseness, precision, and coherence of the generated content 3. Additionally, they plan to incorporate user feedback and A/B testing to refine their models further. Keith also highlights the challenges of determining what to build or wait on in the evolving AI landscape, stressing the need for robust data syncing and enterprise-level features 4.

We have to like the stuff that we produce before we ask our users or before we ask it externally.

--- Keith Peiris

This meticulous approach ensures that Tome's AI models meet high standards of quality and relevance.

   

Overcoming Challenges

Overcoming AI challenges at Tome involves innovative strategies and a user-centric approach. Keith discusses the initial market-centric strategy, focusing on gaining paid subscribers and managing compute costs effectively. He notes the downward pressure on costs due to advancements in model efficiency and hardware 5. Additionally, Keith describes the evolution of AI usage from creating impressive demos to becoming a practical tool that assists users in their work, helping them overcome writer's block and structure their ideas 6.

AI is actually just like a reasonable person who's available 24/7 who can like, you know, iteratively riff with you to help you get your work and your thoughts to a really good place.

--- Keith Peiris

This approach ensures that AI serves as a valuable assistant rather than a replacement for human effort.