Case Study

A virtual assistant for physical branches

A hologram that welcomes customers like a real bank employee

A new ways of interacting

Digitisation has had a significant impact on traditional communication paradigms, leading to profound changes in interactions between people, as well as between companies and customers.

Machine learning techniques have currently reached such a level of maturity that it is now possible to design and develop innovative solutions which, on the one hand, allow the customer experience to be dramatically improved, simplifying it and reducing waiting times, and on the other hand, automate all company services that would otherwise require the intervention of a natural person.

A holographic branch assistant

The solution consisted of a hologram capable of welcoming customers in a physical branch.

Machine Learning Reply was involved in the creation of a virtual assistant for a leading banking company. The solution consisted of a hologram capable of welcoming customers in a physical branch. Thanks to Artificial intelligence techniques, the assistant helps customers with banking services in a digital interaction, just like a real bank employee would do in a physical branch.
Indeed, the client’s main objective was to provide a new interactive experience for the end customer, designed to facilitate and automate banking processes such as payments and withdrawals that would otherwise have required the intervention of an employee.

The virtual assistant, consisting of a 3D holographic pyramid, was thus programmed to be able to answer customers’ frequently asked questions, to help provide payment and withdrawal services, to make appointments with financial advisors and to simulate mortgages and bank loans.

The technologies used

Thanks to Machine Learning Reply’s experience in the Machine Learning realm and to its specialisation in computer vision and natural language processing, the company was able to create a virtual assistant whose interaction with the user seems as natural as possible.

Computer Vision

Thanks to the adoption of computer vision techniques, the avatar is able to read the movements of the person with whom it is interacting in the room and understand when the latter wants to interact with it, moving accordingly and initiating dialogue only when required.

Motion Capture

Motion capture techniques were used to develop the animations, creating the avatar’s movements without reproducing them with software, instead relying on sensors to reproduce the movements of a real person.

Speech Recognition

Speech-to-text and text-to-speech techniques were used to capture, translate and reproduce audio signals. Specifically, the Google Speech-to-Text service was adopted for voice recognition and for converting the dialogue to text format in order to enable the processing of the conversational flow, while the Amazon Polly service was adopted as the text-to-speech technology used to convert the conversational assistant’s answers into audio signals.

Natural Language Processing

Is the core component of the virtual assistant, which manages the conversational flow. In this case, a framework was created to serve as Dialogue Manager, managing the conversational flow and integrating with the Natural Language Understanding module, for which Google Dialogflow was selected. The platform utilises Natural Language Processing (NLP) techniques for understanding natural language, making it possible to recognise a user’s intentions and to extract specific entities based on the context.

Hardware Technology

Front and rear cameras for recognising the user, multi-directional microphones to facilitate voice recognition, with on and off systems to prevent the avatar from listening to itself, a hidden projector and a video camera integrated into the holographic pyramid to capture the user’s movements, as well as speakers built into the pyramid and into the polarised LCD screen, enabling the assistant to interact with the user.

Hybrid Architecture

The various components that make up the solution have been integrated into a hybrid architecture that combines the bank’s on-premises services, with on-cloud services provided by Google and Amazon for the machine learning engines, all while being able to maintain a latency time of less than 1 second, thus making the interaction as natural as possible.

The advantages offered by Machine Learning Reply

The banking company was able to count on Reply’s in-depth knowledge and longstanding experience in the Machine Learning domain.

Being able to guarantee a customised service, Machine Learning Reply designed and implemented a chatbot that seamlessly integrated with the customer’s other corporate platforms, thus enabling each process to be managed in a synergetic and centralised manner.


Machine Learning Reply is the Reply Group company specialised in providing artificial intelligence services and solutions, guiding its customers towards a process of digitisation and helping them become more competitive and data-driven thanks to the adoption of Smart Analytics, Machine Learning and Artificial Intelligence technologies. With extensive experience in deep learning, artificial vision, natural language processing and predictive modelling, the company helps its customers to enhance their business, putting highly experienced development teams at their disposal.