33 Nassau Avenue, Brooklyn
Phone: +1 929 552 0055

16 Toft Green, York, YO1 6JT
Phone: +44 1904 949000

80 Hung To Road, Kwun Tong
Phone: +852 3001 5140

Product Roadmap

Data Training / Interaction Methods / Personalization / Data Connectors / Proactivity / Activity Training / Autonomous Actions 

Our goal is to develop solutions that understand human preferences sufficiently to enable them to perform tasks autonomously.  

The first stage involves allowing users’ to request information in a conversational manner such as statistics, dates and values. At this stage users will also be able to make requests for content creation such as marketing copy for social media, websites and presentations.

The next stage involves connecting external applications to enable the solution to monitor users work in real time and make intelligent recommendations that save the user time, improve the quality of their work and allow them to focus on more complex tasks.

In the third stage the solution will be able to understand the users’ preferences sufficiently to enable it to perform tasks autonomously. By using the training data from stage one, the connections to external systems in stage two and developments in AI technology our solution will communicate with humans and other machines, update records and create reports with a minimum amount of human supervision.


Pulls information, generates content and makes recommendations from activity and training data.


Monitors work in multiple systems through data connectors and makes proactive recommendations.


Learns users’ preferences from activity history and performs task autonomously in multiple systems.

Training From Data0%

Data trained from websites and documents and combined with the language model to provide highly relevant and accurate responses to requests. During the training process, a large dataset of text is used to train the model. The dataset consists of input-output pairs, where the model learns to predict the next word or phrase given the preceding context. The training objective is to maximize the likelihood of the correct next word or phrase in each context. This is achieved by adjusting the model’s internal parameters to improve predictions over time. 

Interaction Methods0%

Interaction via text and voice with both available for input and outputs. This is achieved using a deep learning architecture called a transformer neural network. It uses a variant of the transformer model, which is a type of self-attention mechanism that allows the model to process and understand sequences of words effectively. Speech synthesis technology can produce a natural sounding spoken voice which is close to that of a human. 


A personalized experience with preference options available for users to configure. Each user has individual access permissions to training data to comply with a company’s data rules and to provide each user with relevant answers to requests. Additionally, each user’s activity history can be used to inform responses to provide highly relevant context to conversations. Users can configure the style of the conversation by setting creativity, tone and friendliness parameters.  

Data Connectors0%

All external applications can be connected to enable the solution to monitor a user’s work in real time and make intelligent recommendations that save the user time, improve the quality of their work and allow them to focus on more complex tasks. Live data feeds allow users to receive on demand, up to the minute information on items such as sales orders, marketing analytics, personnel changes and financial forecasts. 

Proactive Recommendations0%

The solution can operate proactively to provide the most value to the users. This is achieved by using training data, connections to external applications and the activity history of the users. Basic recommendations and personalized motivation can be generated from training data and activity history to support users and encourage engagement. More advanced suggestions generated by analyzing patterns in user’s actions to spot opportunities that may otherwise be missed.  

Training From Activity0%

User preferences can be extracted from their actions and used along with data training and external applications to make recommendations and perform some tasks autonomously. This training can use a user’s own activity to help make suggestions along with all other user’s activities to learn best practices and educate all users in the best ways of working. Recommendations happen in real time with some action performed preemptively.  

Autonomous Actions0%

When a sufficient understanding of a user’s preferences is available and a good understanding of the whole team’s best practices exists, the AI can perform tasks in multiple systems with minimal input from humans. Most tasks can be performed autonomously with the machine deferring to humans in high stakes decisions or when the probability of a correct decision is low.