KEY PROJECTS & EXPERIMENTATIONS
By assembling multiple state-of-the-art models, we orchestrate a full pipeline that can be catered into performing many valuable business solutions. We aim to infuse businesses with our AI projects to increase productivity, efficiency and drive them to digitally transform.
Using multiple state-of-the-art text representation, information extraction, and ranking models, we have developed a fast and accurate extractive pipeline engine that can, in a short time, produce a domain-specific document high-level feature extraction model.
The data volume most organizations own is skewed towards being unstructured and ambiguous which makes it invaluable for any use. Doc2Obj project was developed to inject a value in unstructured textual data in any format (e.g. PDF) by converting it into a data structure (e.g. Table). This can help organizations to convert text into digital forms or tables, as well as populating databases. The project can also be integrated into a website and accept documents to fill in a website form automatically. Another dimension of applications is integrating Doc2Obj into other components such as Priority Classification, Anomaly Detection, Fraud Detection, Compliance Checking, Decision Automation, and many others.
In our lab, we have developed a question answering model where a short paragraph is given with a question, and an answer will be produced. Although the model is powerful, scaling is a must to create a business value. Hence, we have developed a pipeline that can read thousands of documents to answer a question in less than 2 seconds.
Question answering can intuitively automate a countless amount of day-to-day tasks that, at some point, become repetitive. We have developed an engine where we accept pre-defined unstructured documents or website URLs where they will be our source of facts, and the engine can answer any related questions. This can be used to automate Website FAQ, Customer Inquiries, and can as well be integrated with a chatbot service to answer customer's questions. Ask Noura, in its current demonstrative shape, can answer any question — without a pre-defined source of facts — by analyzing thousands of pages on the internet, Wikipedia, and noura.ai database.