(Re)Building an AI-Ready data universe

If your data platform isn’t clean enough for humans, it will fail with AI agents.

At BlaBlaCar, tracking competitive data on transport involves ingesting terabytes of heterogeneous public and private data streams. Over time, our Competition Universe suffered a lack of trust by our stakeholders, coming from undocumented business rules, fragile pipelines, KPIs drift and massive cloud costs (over €1k per month).

To make our Competition Universe both human-friendly and AI-ready, we have completely re-engineered our data architecture around four fundamental pillars: context-rich documentation, modular modelling, unified governance and data quality assurance.

During this presentation, we will share the tried-and-tested methods that have enabled us to reduce BigQuery running costs by 85% (from €1k per month to €140 per month), eliminate data ambiguity through strict semantic rules, and implement a ‘Double-Agent’ dbt documentation workflow.

The goal is to provide practical lessons learnt from real-world experience, precise code examples and honest insights into the pitfalls encountered along the way.