Voice-quality analytics platform

Flagship self-service product owned across three major versions and five years. Re-architected a dataset that had grown past 2GB on 60 days of data by pushing aggregation into SQL.

The problem

I inherited a first-generation voice-quality dashboard that teams did not fully trust. The KPI logic had never been independently validated, the data sources were aging, and the Power BI dataset had grown past 2GB on only 60 days of data, which made refreshes fragile and capped how much history anyone could look at.

Voice quality is not a soft metric. Drop-call rate, answer-failure rate, and mean opinion score feed directly into engineering priorities and executive reporting, so an unreliable dashboard is worse than no dashboard.

What I did

I started with trust rather than features, developing and validating the KPI logic against an independent team's calculations so the numbers could be defended in a room full of engineers.

Then I attacked the architecture. Instead of letting Power BI carry the aggregation burden, I pushed aggregation down into SQL and partnered with data engineering on a purpose-built curated table.

Across three major versions the product expanded into E911 and PSAP performance, 5G utilization and carrier aggregation, latency, WiFi, satellite connectivity, and device health across multiple brands.

The outcome

The product became the reference source teams reach for rather than one of several competing views, and it stayed in production through five years of platform change underneath it.

Most inherited reporting does not need more charts. It needs someone to prove the numbers are right and then fix the architecture that made them slow.