Analytics¶

SpeakNode captures detailed data about every conversation for analysis and monitoring.

Conversation History¶

Every agent session is recorded with:

• Session metadata — agent, caller, start/end time, duration, status
• Audio recording — full session audio stored in object storage
• Transcription — text transcript of the conversation
• Telemetry spans — detailed OpenTelemetry traces of every step

Session List¶

Browse all conversations with filters:

• By agent
• By status (completed, failed, active)
• By date range
• By caller information

Session Detail¶

For each session you can view:

• Full conversation transcript with turn-by-turn breakdown
• Audio playback with waveform visualization
• Timing metrics per turn (latency, duration)
• Tool calls made during the session
• Error details (if failed)

Telemetry & Tracing¶

The platform uses OpenTelemetry for end-to-end tracing:

User speaks → STT span → LLM span → Tool call span → TTS span → User hears

Each conversation is a trace containing spans for every operation:

• Span name — operation type (STT, LLM inference, tool call, TTS)
• Duration — how long each step took
• Status — success or error
• Attributes — tags and metadata
• Events — detailed events within the span

Telemetry Pipeline¶

Python Worker → OTEL Collector → Kafka → ASP.NET Consumer → Database
                                      → Langfuse (internal observability)

Data Storage¶