Capgemini leverages Privatemode to code confidentially

Capgemini is building a proprietary software development platform that includes AI-based coding assistants. However, for many of Capgemini's clients in regulated industries, cloud-based coding assistants cannot be used, as the source code would be exposed to multiple parties like cloud and model providers. However, deploying on-premises AI infrastructure for individual customer projects is not feasible because of associated costs and lack of scalability.

This created a dilemma: How can Capgemini make use of modern coding assistants for clients in regulated industries, while ensuring that their sensitive code is protected?

Privatemode AI by Edgeless Systems is a confidential computing framework that enables the use of AI, while keeping all input data end-to-end encrypted. Through Privatemode, a variety of models can be used, including powerful coding assistants like Code Llama and Codestral. Capgemini has quickly integrated the Privatemode proxy and the Privatemode API into their development platform, using cloud-based coding assistants with end-to-end confidentiality.

With the integration of Privatemode, Capgemini can now use confidential coding assistants at scale, without requiring local GPU infrastructure. The integration was seamless, thanks to Privatemode's standard API (OpenAI-compatible API). Now, Capgemini can offer regulated a strong alternative, while benefiting from the efficiency gains of modern tooling for software development. Use cases include defense organizations and insurance companies, where the modernization of legacy code is a pressing requirement.

‍

Capgemini has published a blog post on Privatemode AI, including the use for confidential coding assistants in regulated industries. Below, you can see pictures from Capgemini's GenAI Powered Software Engineering Summit, where we jointly presented the use case.