Scale
On a normal day we process tens of transactions per second (low hundreds of thousands of transactions per day), and during peak events like CyberDay we spike to low hundreds of TPS. We design for 99.9%+ availability and have been consistently within that SLO over the last year.
The Challenge
The hardest part was normalizing very different payment flows and failure modes from each processor (Webpay, Transbank, etc.) into a single, predictable contract for the rest of the platform. That included unifying error codes, handling retries/idempotency, dealing with 3DS and installments, and making sure reconciliation and notifications were consistent so that product teams didn't need to know which acquirer was actually used.
Key Contribution (Tokenizer / Vault / PCI)
I co-designed the architecture of our "agnostic vault" layer and the tokenization flow. The idea is that only a very small, isolated component ever touches sensitive card data; all other services see only opaque tokens. The vault stores encrypted PANs using a managed KMS, with strict network segmentation, audited access, and no raw PAN in logs or application traces. This keeps the rest of the microservices out of PCI scope as much as possible, while still letting us route tokens to different processors in a uniform way.