What to build
A working example that instruments a Deepgram voice agent with Prometheus metrics and provides a pre-built Grafana dashboard for production monitoring. The dashboard should display per-turn STT latency, TTS time-to-first-byte, agent response time, conversation duration, error rates, and estimated per-session cost — with configurable alerts for latency spikes and budget thresholds.
Why this matters
Developers deploying voice agents to production need observability to maintain quality and control costs. Without metrics, latency degradation and cost overruns go undetected until users complain. A ready-made Grafana dashboard with Prometheus metrics gives teams instant visibility into their voice agent fleet — the same production monitoring pattern used for web services, now applied to voice AI. This is especially important for teams running multiple concurrent voice agent sessions at scale.
Suggested scope
- Python backend with Deepgram Python SDK and
prometheus_client
- Deepgram Voice Agent API with custom metric instrumentation
- Prometheus for metric collection
- Grafana dashboard JSON (importable) with panels for:
- Per-turn STT latency (p50, p95, p99)
- TTS time-to-first-byte
- Agent response time breakdown (STT + LLM + TTS)
- Active sessions gauge
- Error rate and type distribution
- Estimated cost per session
- Docker Compose with Prometheus + Grafana + sample voice agent
- Alert rules for latency SLA breaches and cost thresholds
Acceptance criteria
Raised by the DX intelligence system.
What to build
A working example that instruments a Deepgram voice agent with Prometheus metrics and provides a pre-built Grafana dashboard for production monitoring. The dashboard should display per-turn STT latency, TTS time-to-first-byte, agent response time, conversation duration, error rates, and estimated per-session cost — with configurable alerts for latency spikes and budget thresholds.
Why this matters
Developers deploying voice agents to production need observability to maintain quality and control costs. Without metrics, latency degradation and cost overruns go undetected until users complain. A ready-made Grafana dashboard with Prometheus metrics gives teams instant visibility into their voice agent fleet — the same production monitoring pattern used for web services, now applied to voice AI. This is especially important for teams running multiple concurrent voice agent sessions at scale.
Suggested scope
prometheus_clientAcceptance criteria
docker-compose up)Raised by the DX intelligence system.