Mondragon Innovation | Humanity at work – ALVINLOGIS: Research and prototyping of a new conversational intelligent logistics virtual assistant
Mondragon Innovation | Humanity at work
Sector: Logistics
Business Case
The management of a freight network depends to a large extent on the work of the traffic assistant, who is in charge of receiving orders and planning shipments. It is a highly demanding task on which the company’s profitability, carbon footprint and maintenance costs depend, and which is currently carried out manually with no help that can minimise the pressure on the auxiliaries. It is from this context that the idea emerges that could be disruptive in the transport and logistics sector: the virtualisation of the role of the traffic assistant by a smart conversational logistics virtual assistant (ALVINLOGIS).
Objectives
The main objective of this project is the research and development of a new, specialised virtual assistant prototype to replace the capabilities and functions of a human traffic assistant. This new virtual assistant will be immersed in the complex multidisciplinary ecosystem of natural language processing, deep learning, artificial intelligence, high-demand computing, data mining, and so forth.
Use case
The project includes two research projects: in the first block of work, an instructional language model (Instruction Tuned LLM) was generated to maintain a conversation with the user and extract the main variables of the transport to be carried out in order, in a second block of work, to optimise resources and routes for which a predictive AI was built with deep learning technologies on neural networks.
Infrastructure
Hybrid
Technology
Automatic or Deep Learning Text mining AI technologies that generate written or spoken language, images or videos (generative AI)
Data
To train the models we used our own business datasets composed of geo-referenced data from the operation of transport networks and end-to-end supply chain networks.
Resources
Both SmartDataServices staff and internal resources participated in the project, facilitating the assimilation of new data technologies in the know-how of the cooperative. Sixteen people participated in the project (8 senior engineers and 8 technical engineers) and it was carried out in the company’s own cloud.
Difficulties and learning
Among the main difficulties encountered that presented a challenge were: The great difficulty of syntactic-semantic interpretation in multi-contextual environments of transport for corpus development for the NLP algorithm. Limitations of the neural network processing hardware, which made it necessary to modify the architecture.
KPIs (business impact and metrics of the model)
Response times of the developed system were achieved below those of conventional systems (8%). The calculation of ETAs was improved by 12%.
Funding
Private
Collaborators, Partners
SmartMonkey Tecnalia
