The tools make the craftsman. It could be a slogan for DeWalt, but actually, it hits home for every company. At In The Pocket, we rely on a set of daily tools to enhance our productivity. For emails, we use Gmail, while Zoom is our go-to platform for meetings. When it comes to design work, we primarily use Figma, and for coding, Visual Studio Code is our preferred choice. Although there are exceptions, these tools form the backbone of our workflow and enable us to work efficiently. And a little while ago, we added "ChatITP" to our list.

Got time for an origin story?

Exploring our possibilities

Not that long ago, a little thing called ChatGPT happened, comprising GPT3, GPT3.5, and GPT4. Their successive releases have unveiled impressive demos showcasing text completion, text manipulation, code generation, code explanation, and much more. Naturally, these advancements offered us opportunities to optimise our workflow and develop new services for our clients. 

But with all these technologies being brand new and uncharted territory, how did we approach this integration into our toolbox?

In essence, we focused on 3 goals: 

1. Exploration: We aimed to delve into the capabilities of this technology, understanding its strengths and weaknesses. We sought to identify what it excels at, where it requires validation for the final solution, and how it can enhance our work.
2. Enablement: We strived to ensure that Generative AI isn't confined to a single research team but rather becomes an integral part of our entire organisation. We aimed to empower every individual who can benefit from this technology, enabling them to access and explore it to optimize their workflow.
3. Safety: We prioritised the responsible use of this technology at ITP. Given the sensitive nature of the data we handle, we established clear guidelines on its usage and defined its appropriate applications.

The birth of ChatITP

In response, we created a Slack chatbot called "ChatITP." By leveraging Slack as the platform, we minimized the barriers for everyone at ITP to use the chatbot, considering that Slack is already deeply ingrained in our work processes. Initially, ChatITP served as a wrapper for various APIs and models, including GPT4, DALL·E, Stable diffusion, Whisper, and more. This approach provided a secure and effective alternative to ChatGPT. 

Following its release into the organisation, ChatITP served as a sandbox environment for the machine learning team to experiment with novel ways of interacting with LLMs (Large Language Models) and other Generative AI technologies. For example, we explored how we could utilise an existing knowledge base to improve the efficiency of customer support, conducting tests on our own Confluence knowledge base.

ChatITP in action

Additionally, we explored options beyond the typical chat interface. By leveraging shortcuts and Slack commands, we streamlined frequently used prompts such as "translate," "explain_code," and "transcribe file."

You can watch the video below to see ChatITP in action, as used by our copywriter Michiel:

In the first month following ChatITP's internal release, it was utilised over 4000 times. Although not every colleague used it at the same rate, on average, each employee used ChatITP at least once every workday.

It's incredible to witness how Generative AI has seamlessly integrated into our toolset alongside Gmail, Zoom, and Figma. And we eagerly look forward to discovering new ways in which we can leverage this technology in the future.