Songlifty

Led a system-wide backend migration and built the internal tooling for Songlifty, the online music promotion company helping artists increase their visibility across Spotify, YouTube and social media platforms.

The problem

Two problems at once, which is why they were solved together.

The backend was legacy, and it was carrying a service under real load. Everything that made the business work had accumulated inside it, and everything that needed to change next was harder than it should have been because of where it lived.

Meanwhile the operational side ran on repetition. Promotion work at any volume means the same sequence of steps performed again and again, and a team can only absorb so much of that before throughput is capped by people rather than by demand.

Constraints

Traffic does not pause for a migration. In a quiet system a cutover is a scheduling exercise; in a busy one it is a correctness problem, because the old system keeps accepting work while the new one comes up and for a period both have to be true at once.

The internal tooling had its own constraint: it had to be usable by the people running the operation every day, not only by the engineer who wrote it. Automation nobody trusts gets worked around, and then it is just another system to maintain.

And a good deal of the operational work happened through interfaces with no API behind them. Anything automating those has to drive a real browser, which brings sessions that expire, pages that shift, and jobs that run long enough to fail halfway through.

What I built

The migration, from the legacy backend to its replacement, planned and led so the service kept running throughout rather than being taken down and brought back.

The internal management tooling, in Python — the layer the team operates the business through. The automation underneath uses headless browser drivers with real session handling and long-running jobs built to survive their own failures: resumable rather than restart-from-zero, and observable enough that an operator can see what ran, what did not, and why.

That last property is what decides whether operational automation is trusted or quietly bypassed.

Outcome

A platform moved onto a backend that could carry the next thing the business wanted to do, and an operations team no longer rate-limited by how many times a person could repeat the same sequence.

The automation ground here is the same territory as Building Bots and Scrapers with Python — headless browsers, session handling, and keeping long jobs alive against interfaces that change underneath you.