How we built the app that makes playing tabletop role-play games easy.

As someone who started out as a designer, then became a "life-long developer" I quickly became unsatisfied with the tech offered when learning to play tabletop role-play games (TTRPG) like Dungeons & Dragons.

Flashback: Pre-COVID

The industry is run by non-tech companies who were struggling to embrace modern-day tech (at the time) to provide tools to enhance IRL/hybrid game experiences. Needless to say, everything was lack luster.

Flashback: COVID

One positive change in the industry in response to people being forced to stay inside was the rise of the Virtual Tabletop (VTT) experience. New tech companies like The Foundry were formed, crowdsourced, and quickly filled an obvious gap in the industry. Thanks to the VTT everyone was able to continue playing TTPRGs during this emotionally stressful time.

Today: Post-COVID

Now that people are becoming more comfortable with being social again (and want to), the COVID era has left a gap in the industry for the target of players who still love to play IRL games. You are either stuck with going fully virtual (we don't want) or using tools that don't really enhance the game like character sheet management software (Demiplane, DnDBeyond, etc.).

Welcome Nurl

This is where my company Nurl comes in. We are the app that makes playing tabletop role-play games easy by deconstructing everything that makes games complex and presenting it in a holistic way. We then bring all the tools you need to learn, run, or play a game session and enhance them with TTRPG specific features which result in a true enhanced IRL/hybrid game session.

We do this by using AI as our engine to drive the product on a deep level. What does that really mean? Let me break down the high level points of how and why we built Nurl the way we have.

AI as an engine not a trend

Instead of throwing a GPT in our app and saying "we use AI" like the majority of companies today, we instead leverage multiple models that are trained in a specific task to create enhanced feature sets. What do I mean by this?

• We have a model trained specifically to help players learn rules, etc. without cheating
• We have a model trained to enhance multiple notes features like create "one-click dictionaries"
• We have a model specifically to help GM's (people running the games) plan and prepare for sessions
• We have a model specifically devoted to widgets that enhance game tools for dashboards
• We have a model that changes the user experience via accessibility settings

Hopefully you see a trend but I really want to spotlight that last point.

AI to improve accessibility (a11y)

It is shameful how much of a low priority a11y has been to the tech industry. Today, that is changing, but we could still be doing so much more. Let me explain how Nurl is continuing to break glass ceilings in this area.

Each user has a11y settings that allow them to go as deep as putting the exact number they might be in the Austism spectrum. We then use AI to filter the experience (copy, messages, etc.) to cater to how their brain operates naturally. We are also starting to test and validate how to do this same thing for people who have been diagnosed with ADHD.

This is the power we have to improve the world in a positive way today. I challenge every company to "spend the dollars" to actually make your products fully accessible and not just cross points off in a list given to you by a website.

Nurl will be the first app (possibly in the world) to address a11y on that level and I'm proud to be a forerunner of it. Hopefully, this will help grow that spark into a flame?

The app

Because Nurl's target is the IRL/hybrid segment, most of our audience are using a laptop or tablet in their game session. Some even still use pen/paper! We launched a prototype MVP last year (which gained 100 users in 48 hours) and learned that our target doesn't really use mobile. That MVP was also a web-app.

So, instead of staying on the web, we decided to leverage the OS level features of the devices being used which led us to Tauri, a Rust-based Electron if you will. As a technical founder, I can tell you this has been a night/day difference. The app is so much faster, reliable, and gives us a set of OS level features that the web just doesn't provide.

We combined that with the power of SolidJS and we have been able to build our SLC at lightning speed.

AI experiments failed

We have tried our best to push the limits of what is possible today for the things we want the app to accomplish. With that has come some failures that I would love to mention. Most notably one in particular.

Failure: extracting data

While AI absolutely has the ability to extract data, depending on your use case - it may not be the best choice. For example, our app integrates with competitor products to help make transitioning to Nurl easier for players. As with all integration features, the reality is you are subject to X's companies data. For our first integration, this data was a unnecessary giant bowl of pasta.

So, instead of wasting time writing code that just won't scale - because you would have to do it for each integration - we trained another model specific to extracting data. This was OK, but not reliable.

The result would end up with us having to measure the length of the data, slice it up, and deliver multiple tasks just to get X fields to conform to our database schema. Works...but takes a long time and adds additional complexity to the code base.

Pivoting back to the "old school way" (just parsing it ourselves) yielded in an experience that was 1k times faster (literally) than using a trained model.

Each day, LLM's get better so I don't think this will be an issue too much longer? Likewise, there are tools that help like LangChain but I typically choose my battles based on the customer experience over the developer experience (I feel like most companies are backwards here).

LangChain wasn't going to make the LLM faster, only just add a slightly better developer experience while still adding more complexity to the codebase vs. parsing it natively.

AI in the TTRPG community

Like most communities today, there is a segment of people in the TTRPG community who are afraid of AI and the result of what could happen if it's abused - like creators being replaced. This is a topic for another blog that I would love to dive into and address our strategy and how we are approaching it.

Thanks for reading and I hope this helped provide some "insight" and "inspiration" to building AI apps that solve problems - not just making another GPT bot (that's already been done...a lot).

Join the waitlist today and gain advantage on your IRL/hybrid game sessions today at https://nurl.website.