Training Models at Tome: How Engineering Leaders Push the Limits of Machine Learning & AI

As AI becomes a more common professional tool, how can engineering teams make AI even more useful within the workplace? Tome’s engineering leaders had this top of mind as their teams explored new ways to give millions high-quality AI experiences, which led to generative presentations V2.

We sat down with our head of AI, Ves Stoyanov, Archana Sankaranarayanan, engineering director, product engineering, and Sarah Sachs, engineering director, machine learning and infrastructure, to learn more about the tech behind generative presentations V2 and how it’s pushing the limits of generative AI at Tome.

Tome: What were the primary elements in our original generative presentation experience that the team wanted to update?

Ves Stoyanov: There were a couple key elements that were top of mind when we set out to revamp generative presentation. First, we knew it was important to deepen interactive editing and advance the capabilities of the toolbar.

Second, we wanted to give users more control in building their presentations by giving multiple ways to direct page information and design. People now have additional control and versatility in their narrative output.

Tome: How was machine learning used to accomplish this?

Archana S: We took a more advanced approach to working with LLMs (large language models). We needed to help train the model on the fundamentals of what powerful expression and compelling communication looks like.

We also thought about how to separate the actual presentation content creation from how you present the content. What are different presentation layouts that would be compelling? What combination of tiles and objects can we put together in the presentation structure that would enable better storytelling?

Asking for a range of strong approaches to page creation and content creation instead of relying on one deterministic information structure was vital.

Tome: Let’s zoom out a bit from generative presentations. Why is generative AI and machine learning at Tome particularly intriguing and key today?

Sarah Sachs: What’s really exciting is that both are core to the entire infrastructure here at Tome. We are building things where every feature can be ingested by our AI systems, and be easily maintained. That is rare. There are many legacy institutions out there where to build like we do, it's just harder. Tome is uniquely positioned to leverage AI and machine learning at any point of the creation and honing process.

Stoyanov: We’re pushing the limits of LLM engineering and simultaneously deepening the work on our own models. We’re powering pages, and soon will be powering presentation generation with our own models. That’s an opportunity to further hone and refine narrative outputs, which can make the AI experience for people faster, more reliable and more seamless.

We also use a proprietary structure to communicate with the LLM—we call this format, TIF. That's a way for us to communicate narrative structures back and forth between the LLM and Tome’s back end.

Archana S: Generative AI and machine learning is fundamentally changing how humans interact with computers. The opportunity to continue expanding, exploring and pushing the boundaries of what we can do with engineering and in Tome, because it’s an AI-native platform, is incredibly exciting.

As both Ves and Sarah have mentioned, building the tech for even more interactive and iterative co-creation at Tome is a very unique opportunity. Generative creation in the tool is just another illustration of why it’s so powerful to create software fundamentally rooted in AI and ML, where all our systems can understand and converse with LLMs.

Tome: Thank you all for taking the time to share your thoughts with us. We’re excited to see what’s next from your teams!

For more tips on how to create your best work with generative presentations V2, check out our guide and be sure to bookmark the Blog for more of our getting started with Tome 101 guides, product news and more.

You can also get additional thoughts from Sarah and Archana on all things AI and engineering here at Tome.

As always, drop us a line with any questions at team@tome.app. We love hearing from you.