Knowledgator

Whether during creating market reports in a venture firm or collecting target information about chemical compounds in academia, researchers, analysts, and data scientists face routine manual tasks. Almost all data processing activities end up in tabular database construction. A scientist would create a table with columns like: “Compound”, “Molecular target”, “Mechanism of action”, and “Cell line.” A VC analyst would construct a database containing values, such as “Startup”, “Funding Stage”, “Money raised”, “Investors”, “Industry”, etc. Many other examples from numerous industries can be outlined, but the logic stays the same. We develop an Information extraction AI converting unstructured text into self-editing dynamical databases. Such a clever AI solution won’t replace employees but will save their time by auto-filling cells in tables with information extracted from reports, articles, websites, etc. Automating routine database editing will free their energy for professional intellectual work. The system is equipped with zero-shot Named Entity Recognition, Relation Extraction, multi-label Text Classification with probability scoring, and, most importantly, tabular information extraction technologies that cover 100% of any NLP pipeline. Our no-code platform enables users to present a system just a few tens of training examples and fine-tune a model in one click. Users can use default model APIʼs with 83% precision or efficiently fine-tune them and integrate our NLP solutions into their data pipelines. Non-generative AI approach and narrow specification in relation extraction make our solution more accurate and much cheaper compared to GPT4-like LLMs