Say Hello to DiGi: RapDev’s Open-source GenAI Developer

The ServiceNow ecosystem has a new resident! Meet DiGi, your digital wing-bot, always ready to address and assist with the custom needs of your organization.

Back in February 2023, ChatGPT3 made waves as Generative AI hit the market and began one of the most profound disruption events in the history of our industry. Soon after, Google released Bard and from there thousands of other Large Language Models (LLMs) have popped up all over the industry. Every vendor and every keynote speaker now talks about how their product supports GenAI and how it will “change everything.”

‍

So, how many companies are truly leveraging GenAI in their current workplace environment? What value are they realizing?

This has been an ongoing conversation about GenAI. The technology is amazing, but companies are having a hard time figuring out how to make it valuable to implement. What use cases can they drive that will truly move the needle from a value perspective? Ultimately, all of the value that can be driven by GenAI comes down to saving time, saving costs, and reducing risk.

RapDev saw early on the benefits that could be realized in a ServiceNow environment and built a GenAI code-assist product called “DiGi” in June 2023. It was originally designed as a Code-Assist Co-Pilot for ServiceNow Platform Developers. It has the ability to write functions and help generate code for NOW developers to accelerate their time to value by 50% or more. For example, if I wanted to write a function in a script that would validate a phone number field based on RegEx, I may not know how to write that in JavaScipt. So, I can just type the English language of: “@digi Write a function that validates a phone number using RegEx” and DiGi will automatically leverage any LLM you want to provide the code response into your script. You can then quickly validate the functionality and implement the change.

This example would save me hours of effort and work to go look up the formatting or code examples from other developers. Whereas here, I was able to just ask DiGi for help and it provided the answer for me in seconds. If you start to expand this across multiple changes and multiple projects and multiple teams, you can see the immediate OPEX savings as well as be able to get more features to market faster.

We have also built additional functionality into the DiGi platform to drive a number of other value drivers, which we will outline below.

‍

Does it make mistakes though?

Of course it can! Same as any human developer will. You can optimize this in multiple ways and we have been building to make sure we can provide accurate code that will function correctly in Production.

• Multiple Language Models: One way we do this is by supporting multiple language models. Some language models perform better than others when developing code. For example, ServiceNow invested with nVidia to train a HuggingFace LLM with actual NOW platform data and glide development data. This makes their HuggingFace Starcoder LLM roughly 30% more accurate on writing NOW glide code.
• Automated Testing and Validation: Another way to ensure the code is functional and correct is to test it, the same way a developer needs to test their hand-written code. You can automate this testing to run as part of a workflow triggered by accepting the code that was produced. If the test passes, you have good (enough) code. A seasoned developer may be able to refactor that code for better performance or to be more efficient, sure, but functional code is many times good enough.
• Code Comparison: A more advanced version of DiGi that we are currently working on is able to query multiple language models and do a comparison of the answers we receive to land on a higher-probability solution that is more likely to be the best answer possible. Imagine this like an advanced “peer programming” solution leveraging AI. Ask 3 developers how to write a function and you will most likely get 3 different answers. So, we are working on having DiGi ask 3 different LLMs that have been trained on coding and then compare those to get a single, optimized code solution.

‍

Is DiGI just a co-pilot solution for NOW? (“just”… sheeeesh. As if that’s not incredible enough!)

We have already expanded the functionality of DiGi to include a number of pre-built “Actions” that can be performed leveraging GenAI. These all drive value by saving time, reducing costs, and improving the functionality of your ServiceNow platform. Some examples of additional actions that we have built into the platform:

• Build a workflow
• Work a Ticket
• Create new records
• Analyze an Incident to provide real-time context in activity/work notes
• Incident Summary and Troubleshooting Advice
• Document problems directly to KB articles for future reference

And, we integrate DiGi into every Major Incident Management (MIM) solution we build for clients to help them quickly summarize an incident based on all the data gathered, provide troubleshooting ideas that correlate to that summarization, and include that via ChatOps to the Ops teams so they can quickly see what is wrong and have some ideas on where to start to solve the issue. This greatly reduces the MTTR for those Ops teams and helps them identify root cause and rapidly remediate Production issues.

‍

Is this competitive to ServiceNow’s GenAI controller?

The short answer is - Yes and No. We are an elite ServiceNow partner and would never compete against the platform. We wrote this and brought it to market in June 2023, about 3 months before ServiceNow released their GenAI controller in GA (Sept 2023). Much of the functionality is similar and we will discuss this in a later blog, but there are some slight differences. ServiceNow spent $Millions to train their own HuggingFace Starcoder LLM and it’s impressive. While DiGi can support any LLM that’s available to you. We fully intend to merge functionality of DiGi into the ServiceNow GenAI controller on the platform if possible. The key takeaway for folks should be that we at RapDev have been innovating in this GenAI space since June and have driven real value for a number of our clients already. We can help you understand how to leverage GenAI in your NOW ecosystem, no matter which solution you implement.

‍

How do I try this out in my own instance?

This is an open-source product that RapDev brought to market. We want others to be able to work with us to continually improve the capabilities of DiGi. You can check it out on our GitHub. If you ever need help installing it, configuring it, setting up LLMs or anything else,reach out to us at RapDev and we can help you get started towards realizing real value with Generative AI.