# Voice Agent observability with LangSmith 🌟

## Executive summary

This session demonstrates how to achieve deep observability for voice agents using LangSmith when integrating Google's Gemini Live model and the Google ADK. Since Gemini Live is a native speech-to-speech model that bypasses text transcription to maintain low latency, robust tracing is critical. The process involves setting up tools (like a weather assistant), recording both user and agent audio, and utilizing LangSmith to view comprehensive traces that include transcripts, tool calls, interruption events, full cost breakdowns, and even audio playback for debugging.

## Key takeaways

- Gemini Live's Low-Latency Advantage: Gemini Live is Google's native speech-to-speech model; it takes audio directly as input and produces audio output without transcribing to text, which keeps latency low and ensures a natural voice experience.
- Comprehensive Voice Agent Tracing: LangSmith provides visibility into the agent's internal workings, capturing not only standard transcripts and tool calls but also specific events like interruptions and detailed token-level cost breakdowns.
- Production Readiness Tools: The observability provided by LangSmith allows engineers to perform standard LLM operations—such as running evals, adding traces to data sets, building dashboards, and debugging—on complex voice agent interactions.

## Technical details

- Voice Agent Architecture: The demonstration involves building a terminal-based weather assistant using the Google ADK and Gemini Live. The process includes defining and registering a LangSmith Google ADK plugin on the runner.
- Trace Data Capture: The system is configured to record user and agent audio, ensuring that the trace captures exactly what the user heard. The resulting LangSmith trace includes transcripts, tool call records, interruption events, and cost breakdowns.
- LangSmith Functionality: Engineers can leverage LangSmith to debug voice agents by reviewing the full trace, which supports standard text-based features like running evals, managing annotation queues, and sharing detailed performance metrics.

## Practical implications

- Build engineers can safely move voice agents to production by gaining granular visibility into the entire interaction lifecycle, especially regarding latency and cost.
- The ability to capture audio alongside transcripts allows for debugging complex conversational flows that are difficult to track using text-only logs.
- Integrating LangSmith enables standardized MLOps practices (e.g., running evals) on multimodal AI systems.

## Topics

Voice Agents, LLM Observability, Gemini Live, LangChain/LangSmith, Google ADK, Build Engineering, Gemini Live API docs, Google ADK docs, LangSmith docs

Source: https://www.youtube.com/watch?v=bo4ebVkhp28