Cartier: from audit to in-house private LLM.

Cartier. Jewellery and watchmaking house, part of Richemont Group. A continuous tech partnership, from the first audit in late 2021 to today's private LLM infrastructure. A trajectory no sales pitch can manufacture: it gets built one project after the other.

Context

NDA signed in December 2021. First technical engagement in January 2022: codebase access, audit, first quote. Several years later, Abbeal operates on the data + AI stack of one of the world's most iconic brands. Not by luck: by methodical sequencing of projects where quality justified the next one.

The trajectory Compass -> Mapper -> ETL -> Private LLM

1. Compass (audits 2023 + 2025): web app for pricing and benchmarking by marketing teams. Front audit in 2023 then Back audit in 2025 (14 P1/P2/P3 recommendations on security, performance, quality)
2. Mapper V1 + 1.1 + RSP/WWP (2023): product mapping generator for watchmaking then jewellery marketing. 77 days of initial development then iterative extensions
3. Jewellery / Watchmaking Data ETL (2023-2024): competitive data integration pipeline on Cartier's data stack (BigQuery, Dataflow)
4. POC LLM Web Dev (May 2023): first exploratory brick on LLMs front-side. Three years later, it became a structuring project of Cartier's AI infra
5. Private LLM (2026, ongoing): data pipeline -> fine-tuning on Cartier infra -> performance evaluation -> production monitoring. Driver: absolute confidentiality, luxury data never leaves the internal infrastructure

The stack we operate

• Cloud: Google Cloud Platform (BigQuery, Cloud Run, Cloud Functions, Dataflow, Cloud Storage, Cloud Logging) + Firebase
• Backend: Python, FastAPI, Uvicorn, Pandas, Jupyter for data science notebooks
• Frontend: Next.js + React + D3 for dataviz, Vue.js / Nuxt on some internal tools
• Data engineering: custom ETL on BigQuery, dbt-style transformations, orchestration via Dataflow
• GenAI: private LLMs fine-tuned on Cartier infra, RAG on internal corpus, perf eval and continuous production monitoring
• Security: Firebase Auth, SAML/OIDC SSO, parameterized queries (following audit recommendations)

Why this duration, with this client

• Multi-project, not one big build: Compass, Mapper, ETL, POC LLM, private LLM. Each deliverable justified the next
• Multi-function: product marketing, pricing marketing, data office, and now AI. Abbeal isn't filed in a single silo
• Multi-country: Geneva (Cartier Suisse, contracting entity), Paris, Tokyo (direct exchanges with Japan team)
• Human continuity: stable pilot team, skills transfer when scope shifts, never a break in accumulated knowledge
• Premium studio posture: we don't staff in volume time-and-materials. We ship product, we work quality, we document. That's what the Compass -> private LLM journey validates

What this engagement says about how we work

A continuous relationship with a brand like Cartier doesn't get decided at signature. It gets earned deliverable after deliverable. The first time you're judged on an audit. The second on a build. The third on data. The fourth on sovereign AI. At each step, it's the same team and the same demand, just a widening scope. That's exactly the posture we aim to extend with other luxury houses in Europe and Japan.