Iris.ai was founded at Singularity University in the summer of 2015. The project stemmed from a pure love for and understanding of the importance of science, and a belief that if one human being could read and understand every piece of research in the world, we would solve a lot of problems on the spot. The company thus formed around the vision to build an AI researcher to help us achieve this, and that remains our goal to this day. Per late 2019 we offer a number of research assistance tools to academics and industrial researchers, all related to machine understanding of scientific text.
We are a highly distributed international team of overachievers and weirdos. We’re a team of 15 driven, smart and passionate people (if we may say so ourselves) – we are also all a little eclectic and really geeky. We speak 14 languages fluently, live in 9 different countries and count more than 40% women.
We never take the easy way out, mainly because we let societal impact drive our strategic decisions. And while ‘making the world better’ sounds like such a cliché, we try to do that – one happy scientist at a time.
We are developing a science assistant tool for research exploration, navigation and easier comprehension that can machine-read all of the world's research articles and help connect their scientific knowledge, overcoming the current keyword limitations and citation biases. We are building AI algorithms including but not limited to keyword extraction, topic modeling, word-to-word networks building (hypotheses), word-network to text generation, entity recognition and disambiguation, encoder-decoder networks and knowledge graphs for additional resources interpretation and knowledge injection.
The Explore tool starts with addressing keyword extraction, contextual synonyms identification, topic modeling, and document similarity to allow a user to broadly and visually explore the research from a connected database around a given self-written problem statement or 300-500 words. The tool is based on our unique approach towards word embeddings and the research progress we have achieved in context disambiguation and incorporating out-of-vocabulary words, together with our proprietary Word Importance based Document Similarity Metric (WISDM), recently published, achieving state of the art results without compromising accuracy.
The Focus tool leverages similar technologies as the Explore tool and here we semi-automate the Systematic Mapping Study, where the process includes digesting a scientific literature corpus (acquired through conventional search or pre-built digital library) and iteratively narrow it down in a human-machine collaboration to a precise reading list. Our recent advancement of our Hierarchical Topic models will significantly improve the capabilities of this tool.