Over Assistant, the Plug and Play AI avatar

What is the OVRAssistant

The Over Assistant is a 3D realistic human avatar that use artificial intelligence, speech recognition, and speech synthesis to give geospatial contextual information with a neural network trained on Wikipedia corpus and a programmable knowledge managed by the OVRLand owner.

The way the Over User can interact with the Avatar is by voice. Some examples of the Over Assistant interactions are: answering questions, guiding the user to POI (Point of interest) near him, moving on the OVRLand, changing his position based on the Over User activity, inviting the Over User to follow him, changing face expressions, point in a direction, navigate a user inside a shop.

The Over Assistant appears superimposed on the current view of the real world like the augmented reality experiences and can coexist with the Over Contents anchored to the OVRLand.

Over Virtual Assistant Natural Language Processing (NLP) technologies

The virtual assistant will be powered by text to speech and speech to text algorithms, interactions will be by spoken natural language, no typing will be required. Virtual Assistant’s intelligence will be powered by an ensemble of different NLP technologies that can be used both independently or jointly:

Tree search model

This is the simplest and most common NLP technique used to power a virtual assistant, questions and answers are structured inside a nested tree structure and the Virtual assistant exploits such hardcoded knowledge to interact with the user.

Local pattern recognition Question Answering System

Over uses state of the art NLP algorithms built on BERT, a deep neural network developed by Google AI. BERT is the first deeply bidirectional, unsupervised language representation, pre-trained using only a plain text corpus on the whole English Wikipedia corpus. Pre-trained representations can either be context-free or contextual, and contextual representations can further be unidirectional or bidirectional.

Context-free models such as word2vec or GloVe generate a single word embedding representation for each word in the vocabulary. For example, the word “bank” would have the same context-free representation in “bank account” and “bank of the river.” Contextual models - like the one used by Over - instead generate a representation of each word that is based on the other words in the sentence. Such an architecture reaches super-human performances on open Q&A tasks. An example of BERT’s performance on the open Q&A benchmark dataset for NLP developed by Stanford University:

Over AI Virtual Assistant will be Plug & Play, Virtual Assistant’s interactions can be fully customized by simply uploading a text file containing the knowledge that needs to be delivered in the specific circumstance. Both the Text to Speech and the NLP technologies are already functional on the MVP app. The features will be under continuous development in order to improve ease of configuration by the owner and vastness of possible interactions between the Virtual Assistant and the final user.