AI LinguAnalysis LTD

We have recently had great success in applying our research in conversational agents to solve real world problems. The idea behind the new product of our start-up is to implement a new virtual assistant model that allows friendly and more natural interaction rather than the too formal ways of communication in the other existing commercial products. The main novelty of our product resides in the use of new modalities in addition to text or audio in order to improve the experience of the customer, such as images and videos.

Our technology is designed to access millions of unstructured data records from different sources, filter it, and extract the valuable information in real time to enable the ideal answer or task needed to help the customer. The system is automatically improving itself by learning from the experience of the interaction with the user.

This is a big step in the usability of these assistants. Normally the assistant resolves small tasks alarm, food list and call/message someone. For example, our assistant learns the customer way of answering and could be habilitated to answer directly to email or chats. This could be very helpful for the customer because you can freely the time of the people from the small and annoying tasks.

What is the problem you are trying to solve?
The value proposition of this product is focussed on minimising the administrative or repeated tasks our customers are doing daily and that can be automated via our new personalised assistant. This service will allow each customer to acquire a personal assistant that can understand and help with all daily tasks.

Our goal is to give more time to the people, they spend time managing their life without living really. Many of the common task of the day could be automated and our idea is to give a “secretary” to each of you

How did you go about it?
We started from the known virtual assistants and made a list of their features and what they are lacking to achieve the goal we have for our new service. The idea of this value proposition comes originally from our own experience with conversational systems like Alexa and Siri and our daily life activities that can be automated using AI technology.

What technologies did you use and why?
This product aims to design intelligent conversational agents to enable interactive communication with People in different situations and scenarios based on self learning. The second main feature is the integration of the emotion detection model that enables the main goal and novelty of the project: more natural conversation.

How difficult was it?                                                                                                                                 Natural real-time conversations are challenging due to the limited amount of data available to train the models. Understanding the linguistic, contextual features of conversations is an important part of personal assistant systems. There are limitations and challenges in implementing this technology since the textual and visual data is insufficient. In our architecture, a pipeline consisting of CNN and RCNN models implemented to extract textual and visual features and information.

What challenges did you need to overcome?
The big challenge is the integration of different AI models trained on different data types and formats: text, audio and videos.

What was the result?
At this stage the result is a network of AI models that interact with the customers only in English. The actual version understands video audio and text and gives back a context , the next step is to create the “manager” of this context and give the response to the user. The system is still in its Beta version and we expect to start the feasibility study on real users soon.