Real Time Voice Workflows with Zero Instrumentation Observability

About aiOla

aiOla is a voice‑agentic workflow company that enables enterprises to embed real‑time voice driven experiences directly into their applications. Using an SDK integrated into  customer facing products, aiOla captures and processes voice interactions, orchestrates agentic workflows, and delivers near‑real‑time responses that align with strict SLAs while preserving a consistent, high quality user experience.

At the heart of the platform is aiOla’s voice API, built on an event‑driven architecture powered by Kafka and a Kubernetes‑based microservices backend. The system relies on shared GPUs, AWS managed services (including RDS and S3), and a sophisticated LLM stack which primarily uses aiOla’s own models to deliver high‑performance, low‑latency voice interactions.

For aiOla, observability is mission‑critical. Because their SDK is embedded inside customer applications, they don’t always know how, when, or at what scale their APIs are being used. Visibility into real‑time behavior, performance, and failures is essential not just for troubleshooting, but for protecting customer experience and contractual SLAs.

The Problem: Logs without context, and no tracing

Before groundcover, aiOla relied on Coralogix as their primary observability stack. Logs, infrastructure monitoring, RUM, and AWS (CloudWatch) metrics were all centralized there - but there was a major gap.

They had no tracing or APM in place.

Instrumenting tracing into their SDK was a non‑starter. The R&D organization was unwilling to take on the engineering effort, risk, and ongoing maintenance required to manually instrument an SDK that lives inside customer environments.

This created several acute challenges:

• Limited data correlation: Engineers had to sift through massive volumes of logs with little meaningful correlation. Logs, metrics, and infrastructure signals were siloed, with no end‑to‑end visibility across requests.
• Insufficient metrics: Metrics were just enough to keep systems running, but not enough to accurately describe Kubernetes health or application behavior.Application level metrics, such as RED signals, were unavailable, making it difficult to identify systemic issues.
• Compliance and cost concerns: Data egress fees continue to grow, and storing observability data in a third‑party SaaS cloud became increasingly problematic. Enterprise customers required clearer guarantees around data residency and data processing.
  

"The seamless merge of K8s infrastructure monitoring data and the application level observability from eBPF provide insane value, with zero engineering effort.”

- Roy Sela, VP Platform Engineering, aiOla

aiOla knew they needed to move from just logs to logs and traces, and to detect issues proactively through metrics - but without requiring SDK instrumentation.

That led them to a clear technical requirement: eBPF‑based observability.

Why groundcover?

aiOla evaluated vendors specifically through the lens of zero‑instrumentation APM. They already understood the organizational burden and long‑term risk of SDK‑level instrumentation, and believed eBPF was the right architectural direction.

“What you do with the eBPF data is the best we’ve seen.”

- Roy Sela, VP Platform Engineering, aiOla

The Impact

Today, groundcover is deeply embedded in aiOla’s engineering workflow.

The results are already clear:

1. Widespread adoption: The team saw significantly broader adoption of groundcover across R&D compared to the previous observability stack. Developers can understand Kubernetes behavior without relying on DevOps making troubleshooting faster, more independent, and more precise
2. From logs to traces: The introduction of tracing changed how engineers debug production issues. Developers now spend far more time in traces and far less time in logs, which created a more efficient, higher granularity troubleshooting process.
3. Improved product SLA visibility: Zero‑effort dashboards automatically surfaced critical metrics which allowed teams to detect issues earlier, with better context. For a platform where near‑real‑time response times are critical to customer SLAs, this shift directly translates to better reliability, faster incident resolution, and a stronger customer experience.
   

"The adoption of groundcover by R&D is incomparable to our previous stack. It make K8s troubleshooting accessible to them in ways they didn’t have before.״

- Roy Sela, VP Platform Engineering, aiOla