Iris.AI version 2.0 is here!
We are thrilled and excited to announce that Iris.AI v2.0, the newest version of Iris.AI, is live!
When we started defining Iris.AI in October 2015, we knew what we had set out to do would be complicated and take time. We agreed that we wanted to launch something within 3 months, and so the 1.0 version (as we know it) was born. But of course, this was only the tiny beginning and since then we’ve been heads down; building, fundraising, discussing, talking to users, laughing, and no startup dream would be complete without a lot of swearing and occasionally crying.
We expanded our tech team with a few more amazing people. We built an AI trainer platform and community with hundreds of people helping Iris.AI learn. We started a collaboration with a great design firm. We found some great people who believe in us. We found active and eager pilot customers. We evaluated algorithms, libraries, approaches, methodologies and tools. We made hard choices and prioritized. And we coded. We coded A LOT.
And here she is! Our baby Iris.AI, our labor of love, has become a toddler. Still with much to learn, but this time we believe she has something to teach us, too. Let us introduce you to the changes you’ll find when you start exploring.
- Brand new UX. While you’ll recognize the functionalities and visual elements, the overall design of Iris.AI has seen a major upgrade – and is the instantly most visible change.
- Input data – for real this time. While the TED talks were fun and engaging, they were just the tiny beginning. You can now start with any research paper in the world. Iris.AI will work with the abstract and present you with a context-based science map.
- Output data – 15x more. Iris.AI now helps you navigate more than 30 million Open Access research papers.
- New engine under the hood.
- Neural network. New as of this version is that we now do our topic modeling in a neural net. Currently, this has given us about 7% improved results – and it will continue to improve as the tool is being used.
- Supervised learning. With major thanks to a heroic effort to our AI trainers, that have trained Iris.AI on more than 1000 TED talks for us, we are incorporating a layer of this supervised training. This has improved the results with around 5%, for now.
- Opportunities for pilots with custom content. We now have the ability to train Iris.AI on specific scientific content, and have opened up for a limited number of pilot customers to join us.
I could go on forever about how excited we are, but I’d rather you go check Iris.AI out for yourself. And let us know what you think, if you find any bugs or if you have suggestions, will you? (firstname.lastname@example.org)