Avatar PI, not only a Chatbot: the human-like Virtual Assistant for Poste Italiane
Automating corporate processes through an NLP Chatbot
A Virtual Assistant trained to interact with people and support conversations on a range of topics. The creation of an Avatar for Poste Italiane starts with the development of an AI Assistant Machine for internal needs of that Research and Innovation Laboratory. Further prospects lie in the definition of automates tasks architectures for support functions in high complexity contexts. The experimentation is within the scope of Large Language Models for the interpretation and generation of human language.
Un Chatbot con fattezze umane in grado di riconoscere volti e linguaggio naturale
Project name:
Avatar PI
Duration:
11 month
years:
2020-2022
Target markets:
Image Recognition, NLP, LLM
Our solution
After the preliminary definition of a conversation flow on which to train the Virtual Assistant, the main services offered for Chat Bot NLP and natural language recognition were tested. The identified solution is made up of different and interconnected components:
- A front-end that allows the on-screen visualisation of the Virtual Assistant, which is expressive and has human-like features. This interface hosts both the conversation transcript and the visualisation of the retrieved information.
- An NLP Chat Bot to interpret the conversation, classify the research intents and elaborate answers.
- Face Recognition for already registered users in order to customise dialogue with the help of neural networks.
Advantages
The fusion of these aspects in a single application guarantees high levels of spoken language understanding. In addition, it demonstrated pronounced elasticity in dialogue configuration, providing immediate support in a variety of areas. The 24/7 Virtual Assistant proved capable of autonomously interacting with the registered employees, establishing a personal dialogue and performing certain tasks for them:
- Reserve shared corporate space
- Organise meetings
- Search through contacts in the corporate phonebook
- Retrieve information on colleagues’ activities
