Use Case Intelligence RSS Search use cases Search
Mistral Commercial and Professional Services global

Capgemini

Application Refactoring

Capgemini needed a coding assistant that developers on regulated client projects would actually adopt, but public-cloud AI services conflicted with sovereignty rules and earlier open-source assistants were too inaccurate for daily use across modern and legacy codebases.

Read the original story Explore all use cases
100% developer adoption rate across all client projects
50+ client projects reporting productivity improvements in less than 6 months
90% code completion accuracy vs. 50% with previous solutions

Reported impact

Business outcome

Capgemini raised developer adoption from 30 percent to 100 percent, improved code-completion accuracy from about 50 percent to 90 percent, and delivered productivity improvements across more than 50 client projects while meeting sovereignty requirements for regulated sectors.

  • Productivity Improvement
  • Risk Reduction
  • Speed to Market

Solution

Capgemini built RAISE for Private Software Engineering, also known internally as SovBox, as a self-hosted software engineering assistant for regulated client work in aerospace, defense, and the public sector. The deployment kept existing IDE and CI/CD integrations in place while replacing the underlying model with Mistral's Codestral, so teams did not need to re-platform their development environments. Developers use the assistant for code completion and chatbot-style help across modern stacks and legacy languages such as ADA, C, and C++, with deployment methodology and coaching wrapped around the rollout to improve adoption. By keeping the service private and sovereignty-compliant, Capgemini could support projects that could not use public-cloud AI while still improving coding accuracy, workflow continuity, and delivery speed.

Data flow

Developers invoke the private assistant from existing IDE and CI/CD workflows, the self-hosted platform routes coding and chat requests to Codestral, and the responses return inside the same tools without moving regulated client code into a public-cloud service.

Solution architecture

2 components · 2 layers
  1. Serving
    • Codestral Supplies the code-completion and chat model Capgemini runs inside the assistant.
  2. Orchestration
    • RAISE for Private Software Engineering Provides Capgemini's internal private software engineering assistant for regulated client projects.

Architecture clues

  • Capgemini says it is expanding to Codestral Embed and Devstral for RAG and CI/CD use cases.
  • The assistant supports legacy languages including ADA, C, and C++ for applications dating back to the 1980s.
  • The switch to Codestral required no changes to the existing infrastructure.

Evidence from the source

"Accuracy jumped from 50% to 90%"
"Capgemini's transformation came when they transitioned from their previous open-source LLM to Mistral's Codestral in their self-hosted coding assistant solution."
"Developer adoption jumped from 30% to 100%, with all approximately 50 client projects reporting productivity savings."
"The switch required no changes to their existing infrastructure-they simply replaced the underlying model while maintaining other IDE and CI/CD integrations."
"While generative AI's power is typically delivered through cloud-based solutions, these would not comply with all contexts in terms of sovereignty requirements."
Transparency Evidence & methodology 100/100 reference value · 89% confidence · 1 source

Review how this reference was scored and trace each normalized claim back to its documented source.

01

Scoring methodology

How scores are calculated

Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.

Customer reference value100/100
  • Problem clarity20/20
    Measures whether the customer's operational problem is specific enough to compare.
  • Implementation depth25/25
    Rewards named components, data flow, integrations, and architecture detail.
  • Outcome specificity20/20
    Measures whether the reported benefit is concrete rather than a generic claim.
  • Quantified impact25/25
    3 metrics describe measurable impact.
  • Deployment scale10/10
    Rewards evidence of a live or scaled deployment and meaningful integration scope.
Evidence confidence89%
  • Source authority20/25
    The main evidence is an official vendor-published customer story.
  • Claim traceability21/25
    10 of 12 key fields link to a named source.
  • Direct evidence20/20
    5 direct source quotes retained.
  • Date provenance10/10
    The publication date comes from the source page or publisher.
  • Source transparency8/10
    1 source reference document publisher, URL, authority, retrieval, and supported fields.
  • Implementation specificity10/10
    Rewards source-backed components, data flow, quantified impact, and a specific outcome.
02

Traceability

Sources & provenance

Field coverage shows which facts each source supports. Dates are shown only when documented by the source.

  1. Mistral

    Capgemini customer story

    First-party
    https://mistral.ai/customers/capgemini/
    Source type
    Vendor-published customer story
    Published
    Apr 24, 2026
    Supports
    • Customer
    • Business problem
    • Solution
    • Products
    • Solution components
    • Data flow
    • Integrations
    • technical area
    • use case category
    • Business outcome
    • outcome category
    • Metrics
    • Architecture
    • evidence quotes
    • Publication date

Date evidence

  • Published on source page Apr 24, 2026 · Mistral · First-party
    Open date source
Not separately sourced Industry, Region, Company size. Treat these normalized attributes as lower-confidence until a dedicated source is added.

Related

Similar use cases

See all like this
OpenAI Oct 8, 2026 Endava Commercial and Professional Services Global Endava needed to remain relevant in an AI-centric market by redesigning software delivery and cross-functional workflows, not just adding point tools to existing engineering processes. 11,000-person global workforce Application RefactoringIT Operations Automation AI agents · ChatGPT Enterprise · Codex ServiceNow Modified · Jul 30, 2026 Syslabo Commercial and Professional Services Tokyo, Japan Syslabo needed to replace a 30-year-old legacy sales support stack, remove silos between separate sales tools, and give business users a practical way to build and improve internal applications without long specialist… 2x Faster onboarding for App Engine with AI Agents Application RefactoringSales Automation AI Agents · App Engine · Now Assist for Creator ServiceNow Modified · Jul 17, 2026 Genpact Commercial and Professional Services India and Global Genpact wanted an internal ServiceNow development center of excellence that let professional developers and business users build applications faster, but it needed governance, curated AI access, and a review model that… 140,000+ people in many countries Application Refactoring ServiceNow AI Platform · ServiceNow App Engine · ServiceNow Otto