SquadStack Observability Platform: Unifying Voice AI Reliability Monitoring
Designed and shipped a unified observability dashboard to reduce downtime and cut enterprise client onboarding time.
Bhargava T
AI Product Manager at SquadStack.ai
From their time as
AI Product Manager
SquadStack.ai • 2026
Overview
Bhargava inherited a reliability problem at SquadStack that was structural: seven systems come together to make a successful voice AI call, and when something broke at scale, there was no single place to look. Builders were hopping across Grafana and multiple fragmented data sources to diagnose high-latency calls, with no way to tell whether the issue was in STT, LLM, or TTS.
The Story
Bhargava inherited a reliability problem at SquadStack that was structural: seven systems come together to make a successful voice AI call, and when something broke at scale, there was no single place to look. Builders were hopping across Grafana and multiple fragmented data sources to diagnose high-latency calls, with no way to tell whether the issue was in STT, LLM, or TTS.
He approached it in phases. The first phase was visibility: build a unified view that showed builders exactly where calls were failing, so they could take action. The second phase added threshold-based alerting, so builders received a notification when latency spiked above campaign averages rather than discovering it after the fact.
Getting there required working through the engineering team, who had no OLAP layer to support the solution. The analytical infrastructure had to be built from scratch, which took a couple of months. Bhargava kept the builders' daily complaints visible to the engineers throughout, and once campaign volume increased, the engineers committed to the build.
The result was a 95% reliability rate across campaigns and a reduction in client onboarding time from seven weeks to four, driven by the reusable prompt architecture and observability tooling working together.