Context
The media exposure of ChatGPT prompted BpiFrance to consider new use cases to align with evolving consumer expectations, aiming to stay at the forefront of innovation and maximize its chances of maintaining a competitive edge.
The objective was to bring new technical and business knowledge by working on and delivering a Proof of Concept (PoC) that leverages the latest Large Language Models (LLMs) to meet a user need. To achieve this, several use cases around generative AI were studied and prioritized based on the value they bring to Bpifrance, ease of testing and production implementation, and compliance with security and conformity criteria.
Challenge
- Lack of perspective on generative AI technologies and the proliferation of the ecosystem:** This raises questions about technological choices, such as which model to use for which use case and which technological components to integrate, as well as concerns about model maturity, ethics, and responsibility.
- Uncertainties about the preferences and knowledge of potential users regarding these technologies due to their very recent rise:**
- Lack of client knowledge on the specifics and challenges related to Machine Learning:**
Solution
A project carried out in two phases:
- A two-week exploration phase to choose the right generative AI use case and define a technical and data strategy.
- Four two-week iterative phases on a demonstrator (8 weeks in total) to deliver a product usable by 50 people on their personal phones by a specific date.
Results
The chosen use case is a bot that offers relevant services and products in response to a user query while assuring them of the quality of the response provided. It comprises a search engine component and a question-answering component.