Incredibly well positioned, and finally properly funded – Iris.ai

Last week, we were finally able to announce publicly what we’ve worked on finalizing for months; our Series A of almost €8M was closed, doubling the amount of funding we have raised total and placing us thoroughly in territory we have not been in before: Properly funded for the next couple of years. It’s a luxury, it’s a privilege, it’s a commitment. And most importantly, it is an opportunity. 

We knew we were early when we started Iris.ai. Probably too early, but we had the right team, idea and opportunity to start something important – so we went for it. We knew the direction the technological landscape was moving in, although we had a hard time convincing investors of it. We’ve survived on research grants and a few very patient and very trusting angel investors. All the time, we’ve been driven by knowing what we do is unique, valuable and our time would come. We watched patiently as two waves of competitors, struggling to build valuable business models and solid enough technology, came and went. We offered tools to the right segments at the right time, all the while spending a disproportionate amount of our scrappy startup budget on research. 

With the introduction of OpenAI’s GPT models, we’ve seen the third wave of competition rise – this time with an absolute flurry of new entrants to the market. Most basing their tools on the APIs of the big players, most with very limited research departments and most with very little understanding of some vital facts we have learned to the core over the past few years:

1. If you are going to sell AI solutions to corporate R&D, privacy and security needs to be front and center. You do not send data to any third party provider. Any data you store yourself should be the bare minimum and not shared in any way with other clients, including for training of machine learning models. For the bigger clients with more strategic collaborations, you offer private cloud or on-premise deployments. 
2. Hallucinations are obviously a massive problem with the LLMs, and this is extra problematic in the field of science. But it’s not just hallucinations – and fine-tuning of models is not a sufficient approach. A multifaceted and extensive systematic approach to finding, organizing and retaining factual information and data throughout the system is key to success. We’ve spent the last 8 years focused on this, and will spend the next 8 years improving upon it. 
3. Researchers – our beloved users – are skeptical and analytical by nature. That’s why they’re doing the job they do. Which means frameworks for evaluations – of models and of features – are essential tools in our arsenal. Without them, you won’t get far in convincing your users of your platform’s performance. 
4. Science is not just science. Every niche field has its own vocabulary, own expressions and abbreviations and peculiarities of language. We’ve developed a method for automatic domain adaptation which catches all domain specificities, which increases the accuracy of the tools from under to above human accuracy. That’s key to success.

So what this round means to me, is proof. Proof that we are in the right place at the right time. Proof that what we are doing is unique, makes sense and that our world is ready for it. Proof that there is a market – given all our recent competition – but that we have a major head start, both technologically but especially in know-how and market understanding. Now, it’s up to us to prove the next phase: that we can use this unique momentum and opportunity to grow. 

Bring on the next chapter! 

Anita Schjøll Abildgaard

CEO & Co-Founder of Iris.ai