Pythia
Πυθία (Pythia) is our automated Question Answering framework being developed in collaboration with Natural Language Processing specialists from the University of Warwick and Alan Turing Institute. Pythia uses neural transformer models to generate synthetic question-answer pairs from your data in order to fine-tune state-of-the-art question answering models for your use case without the need for long-winded data annotation exercises.
Cielo
Cielo is our state-of-the-art AutoML platform under active development in collaboration with the University of Essex. Cielo can be used to quickly train models to classify numerical or textual data. Cielo plugs right in to our MLOps deployment process allowing us to quickly train, test and deploy new models.
Entrega
Entrega is a one-click-deploy ML model serving framework that integrates with MLFlow - our MLOps platform of choice. Entrega allows us to quickly stand up REST APIs to serve models trained by hand or AutoML solutions (such as Filament Cielo). Entrega supports hot-swapping of MLModels through container-based architecture allowing for A/B testing and continuous training and improvement of ML features.
Telaraña
Telaraña is a framework for simultaneously carrying out named entity recognition and relationship extraction (e.g. knowing that Bob is a person, Filament is a company and that Bob works for Filament.). Telaraña leverages a combination of state-of-the-art named entity recognition and relationship extraction models and uses few-shot-learning to learn new types of entities and relationships without long-winded data annotation exercises.
Chassis
Chassis is our web project boilerplate. It is a batteries-included starting point for modern web projects that provide services via a REST API. Chassis includes out-of-the-box full-stack implementations for authentication and authorisation, email notifications and easy integration points for new use-case-specific functionality.
Argo
Argo is a web data collection framework that can interact intelligently with modern reactive websites in order to extract unstructured information. Argo provides end-to-end natural language processing and enrichment and can capture and store keywords, entities and relationships of interest from almost any web page. Argo is compliant with anti-scraping measures and robots.txt
