Case 2 · Data platform · AI shipped
From SQL gatekeepers to plain-language answers
A 20TB / 10-billion-row analytics platform — and the AI assistant that shipped on top of it, to internal teams and 1,000+ merchants.
The situation. A European payment provider serving e-commerce merchants. Transaction data outgrew the tooling: reporting ran on limited extracts from the transactional database, merchants couldn't see their own business, and non-technical staff needed SQL skills they didn't have. Meanwhile, "AI strategy" existed only as slideware.
The leak. Decisions made on incomplete data; analysts acting as gatekeepers; merchants blind to their own numbers. The data existed — the access, the depth and the interface to use it didn't.
What I did. Led the internal analytical team end to end: OLAP architecture on AWS (S3 / Athena / Glue / Step Functions) sized to 20TB+ with fact tables past 10 billion rows, delivered under the strict cost envelope set for the project. Onboarded additional data sources over time. Launched the merchant self-service analytics product (Kubernetes, downloads + SFTP) now serving 1,000+ merchants. Directed dashboards and monitoring (Metabase, KNIME, Power BI). Shipped the natural-language SQL assistant (Bedrock / Claude) — now extending with MCP connectivity.
You don't need "an AI strategy". You need one tool your team actually uses, on a platform that doesn't buckle. That chain — platform, product, shipped AI layer — is exactly what I build, in payments, under real budget constraints.
Working on something similar?
Twenty-five minutes on a call — or breakfast in Porto or Minho. You describe the stack; I'll tell you what I'd measure first.