Parity’s products help its customers answer business-critical and sometimes life-saving questions, such as:
- Publisher: “From my vast corpus of content, how can I accurately determine which researchers and institutions are the most productive and innovative?”
- Researcher: “Which grant-awarding bodies are most likely to fund my project, and which journals are best suited to publish my results?”
- Geneticist: “Which gene variants are most likely associated with the symptoms displayed by my undiagnosed patient?”
- Pharma Company: “Based on their genome, which patients are most likely to experience good results from our cancer therapies?”
We’re addressing difficult data science and software engineering challenges, for example:
Problems
- Author Disambiguation: Disambiguate authors in a corpus consisting of tens of millions of published articles and requiring billions of comparisons between author instances
-
Predictive Clinical Decision Support:
Analyze clinical data in real time to predict which hospital patients are at risk of sepsis, a life-threatening disease
Data Science Approach
- Rule-based and learned probabilistic graphical model with hundreds of features that make use of all available structured and unstructured article data and links between articles
- Proprietary clinical NLP incorporating multiple learned classification models and concept extractors. Novel machine learning approach to predictive, time-series decision support problems
Engineering Approach
- Cloud-based, massively distributed and multi-threaded Java application enables arbitrary horizontal scalability
-
Demanding throughput, reliability, and HIPAA security requirements met with complex event processing Java core, UIMA NLP pipeline, NoSQL data store, and unique multi-level security architecture
