Tabris A Machine Reading Comprehension (MRC) systemthat finds answers for open-domain natual language questions. Request demo Try it through magi.com Additional Angles to Viewing Questions Tabris has been created to support Magi's revolutionary feature of directly providing answers to users' natural language queries. Unlike optimizing for a typical SQuAD task, the model accentuates two types of results - passages that are relevant to and complete in addressing the question, and phrases that are suitable in logic and succinct in wording.
Ramiel A performance-driven web-scale search enginefeaturing an original indexing structure. Request demo Try it through magi.com Fast response and reliable result quality under monstrous data scale Ramiel is designed to handle data of the scale of the Internet. It is critical for a search engine to have both high response speed and search result quality. With the originally designed indexing structure and algorithm, Ramiel has proven to be a reliable engine.
Leliel An end-to-end neural query understanding systemthat understands natural language queries. Request demo Try it through magi.com Forged by firsthand experience Unlike intact articles, queries in natural language often lack key information or grammar components, which sets substantial obstacles for programs to process. Peak Labs has been handling first-hand user queries for years, and has distilled the experience into Leliel, which now supports nearly every module in Magi project that deals with query inputs.
Ireul A neural open information extraction systemthat understands knowledge expressed through natural language. Request demo Try it through magi.com Exhaustive open information extraction (OIE) Through multiple specifically designed neural networks and hypothesis searching algorithm, Ireul is able to utilize in-depth semantic features. Multilingual with little resource Through multiple specifically designed neural networks and hypothesis searching algorithm, Ireul is able to utilize in-depth semantic features and exhaustively extract all the potential combination of phrases that can form a piece of knowledge, which enables automatic knowledge base construction and numerous business intelligence applications.
Arael An all-in-one multilingual analyzing pipeline. Request demo Try it through magi.com How multi we talk about when we talk about multilingual Arael supports more than 170 languages and can process terabytes of data in the form of stream. For languages like Chinese, Japanese, Korean, Thai, etc., Arael is able to deal with overlapping ambiguity and out-of-vocabulary words/phrases using machine learning. Easy to implement into workflows Not only can Arael segment basic graphemes, words/phrases, and sentences, Arael can also be configured with analyzers and filters to form customized workflows.