Word from the founder: I have now joined ServiceNow - an organization that shares my vision for enterprise AI and is ahead of the curve in terms of AI and Gen AI adoption. As such I shall put this site on-freeze until a new purpose is discovered for it. In the mean time I shall be pursuing my interest in enterprise AI, workflow automation, and Gen AI agents through my role as an AI specialist within ServiceNow.

Technical update: We've decided now would be a good time to migrate our tech stack from Django+HTMX to Fast API+React. It's actually something we've been working our way towards for the past month (the database has already been disentangled from Django). We are also looking into setting up our own LLM instance as a precursor to utilising LLMs hosted within client environments."

Our first product demo is out. In this video we demonstrate our AI Assistant Emma, interepeting a user request against a set of possible instruction sets then performing the task of updating a Powerpoint slide deck by following step by step instructions. It demonstrates how AI Assistants can be controlled to only modify items on a slide deck that the user would like modified, with additional guidance being provided by the slide deck owner.

Good progress this week with getting the Powerpoint feature up and running. Hope to have a video demo ready in the next week or so. Note that our goal is to build an autonomous agent capable of performing multiple tasks, and the slide deck update is one such example of a skillset we are enabling.

This week has been about experimenting with prompt engineering for the Powerpoint skillset - we are teaching our assistant Emma to adapt an existing sales deck to a new customer or product. Two Open AI announcements also caught our attention. We've also been excited to hear about Open AI launching the ability to fine-tune ChatGPT models - we think this will be a game changer for the LLM landscape and something we will be applying to our own project.

We were less excited to hear about the Open AI Enterprise plans, specficially what is mentioned at the end of their blog post. This gives us some pause for thought due to the overlap there with what we are building. Our goal is to complement Open AI not compete with them so their move into serving business users directly will require us to reconsider our approach.

We now have the AI agent linked up to our chat interace - here's a first look!

This week has been about hooking up the new websockets based chat interface with our existing AI assistant. Screenshots to follow on completion!

We have decided to provide regular updates. We began this project back in March and our idea has undergone several evolutions - and likely to evolve further still. Our current goal is to build a multi-purpose AI agent, capable of switching between a number of different complex tasks. We currently have much of our front-end set up and have been testing out different chat interfaces including mobile.

This weeks focus has been on upgrading our chat interface to use Websockets. This will enable much smoother interaction with the user and sets us up to easily incorporate advanced capabilities in the future.