Technology

Having researched the competitive landscape, we found that most EdTech companies are focused on developing and delivering content or creating tools that enable others to create content.

We chose a different approach. Our technology creates content from unordered resources with virtually no human intervention other than uploading or copy and pasting from existing documents. Our computational linguistic strategies then use distributional semantics to generate, answer, and score questions using novel symbolic computing methods to clean up messy ML output using fully proprietary logic programming frameworks for syntactic and semantic processing.

Since 2017, we have been pursuing novel work in syntax theory, the intersection of compositional and distributional semantics, logic programming, and automated inference.

Soffos Visual - LaptopBullet Icon - StreamBullet Icon - Time savingBullet Icon - Learn

While we leverage standard ML techniques (transformer encoder/decoders, vector-space word embeddings, topic clustering, etc.), we are also developing error correction frameworks that target the syntax-semantics interface where large language models usually fail.

Eventually, we hope to expand the techniques we are developing to incorporate taxonomies and ontologies that our customers bring to the table for tasks such as Named-entity recognition (NER) and document classification.

Given that few NLP companies are exploring areas such as logic programming and formal semantics, we must be circumspect about our current technical direction as we work to secure patents and publish both academic papers and technical white papers on our methods.

We have chosen to standardize on processing frameworks that use the most logically and mathematically pure theories of syntax available, to ensure clean translation of natural language into logical relations. This has been one of the core goals in AI since the late 1950s, and one that has proven elusive in terms of standard machine-learning approaches.

In this way, we empower our users to query, access and apply knowledge stored in unstructured text, so they don’t have to rely on keyword search or statistical text-mining tools.

The result is the ability for our AI to present information in natural language, using the conversationally-based Socratic Method of teaching – this strategy ensures that information is fully ingested and retained by the trainee.

We have filed two patents that are pending with the USPTO:

•  Method and Apparatus for Teaching Using a Machine Learning Algorithm.

•  Method and Apparatus for Autonomously Assimilating Content Using a Machine Learning Algorithm.