Introducing the Motley tech blog

What Motley is about

The vision of Motley is to enable number-intensive storytelling, supporting the user in telling the story about their domain that is both supported by data and reflects the user’s priorities.

‍

We will achieve that by finding stories in data, cross-linking them with text information, and using that to generate communication artifacts, such as presentations and reports; eventually also interactive ones that enable the user to drill down on any section to get more detailed information, to any depth they desire.

‍

Of course, Rome wasn’t built in a day, so our initial product focuses on generating number-intensive decks, such as Quarterly Business Reviews used by Customer Success teams, blending traditional (deterministic) automation and AI techniques to combine flexibility with ease of use.

‍

How we get there

While we are heavily using Large Language Models in our product, and often do so in ways that could be described as “agentic”, we don’t place that at the center of our product.

‍

In fact, we believe that the “AI agent” technique is being commoditized fast enough that using one is no more special than using a database. The question is what you use it for.

‍

In our case, Generative AI is mainly used to reduce friction and provide flexibility to the product, while keeping tight control over the validity of output.

‍

Over the last two years of using LLMs to solve a variety of practical tasks, we have developed a certain style of using them that focuses on generating structured objects rather than freeform text or SQL, using tight validation loops. That style is different from what we’re seeing many other people use, and so worth relating.

Another aspect of our style of using GenAI is treating the LLM/”agent” not as the centerpiece of the flow, but rather just as another kind of component in a software/Machine Learning/Data Science toolchain. This has some interesting implications, for example, for how tools used by an agent should pass data to each other, without depending on the agent not mangling it.

‍

Overall, we treat GenAI as just one more tool in our toolbox, used along with the other tools to provide a delightful and powerful user experience, realizing the vision described in the previous section.

‍

What this blog will be about?

Now is a great time to be alive as a software engineer and machine learning/Data Science practitioner. Generative AI techniques make easy many things that were hard or virtually impossible only 3–4 years ago, with promises of more to come.

‍

At the same time, like with any transformative technology, there is an explosion of related techniques, standard, protocols, and products, some of which are massively useful, and others that aren’t even worth the trouble of learning them — and the landscape changes monthly or even weekly.

‍

In this blog, aimed mainly at technical audiences, we will share our journey of building Motley, and our experiences of trying out many of these techniques and products, and whether they work for us or not.

‍

This will include the specific ways we’ve developed of using LLMs reliably as part of bigger workflows, our experiences with using third-party tools that look like they might be useful, and any other nuggets of wisdom we feel might be of interest to fellow geeks.

‍

Is there anything specific you’d like to see us cover? Let us know!