Ashans
Khadka

The $376K annual vendor cost went away once I built the deletion workflow in-house.

A deletion request touches every store that holds customer data. I could not call one complete until the relevant systems had confirmed it. The vendor workflow was expensive and put that legal obligation outside our control, so I built the replacement.

New stores and retention rules kept changing the shape of the work. I kept the business rules separate from storage so a change underneath did not mean rewriting the core decision.

The rate-limited GDPR/CCPA services now run in every production environment. They process approximately 200K PII-based requests per month while preserving data residency and protecting production performance.

What I'd do differently: I would define, validate, and version the data contract on day one. The first version relied too much on one system's document shape, which made later changes harder to reason about.

The MongoDB fleet stopped taking months to upgrade across seven environments.

I led the 4.4 through 8.0 upgrades. I did not want an upgrade to become routine until we had a tested way back, so each change used a maintenance window, a standby-cluster safety net, and a rollback path we had already run.

I wrote reusable runbooks and brought teammates into the work. That cut the upgrade cycle from months to weeks without making the process less safe.

The CDC pipeline could look healthy while data was still missing.

A green dashboard was not enough proof that a change event had reached the lakehouse. I added continuous validation between the events and landed data, so we could see whether the path was actually complete.

It caught five issues before they became incidents: schema-driven lakehouse gaps, source-database ingestion delays, MSK throttling, and oversized events rejected by MSK. I presented the result to the wider engineering organization.