How to Build Real-Time Sales Coaching with the Sonesse API

Introduction

Real-time coaching is transforming how sales teams operate. Instead of reviewing calls after the fact, your reps get instant guidance while they're on the call. With Sonesse's streaming transcription API, building this capability into your product is straightforward.

In this tutorial, we'll build a real-time coaching overlay that analyses conversations as they happen and surfaces actionable coaching prompts.

How It Works

The architecture is simple: Sonesse joins the meeting and streams transcript data via WebSocket. Your application processes each utterance in real-time, runs it through an LLM, and pushes coaching prompts to your frontend.

• Sonesse API handles meeting integration and real-time transcription
• WebSocket connection streams transcript events with sub-second latency
• Your coaching engine analyses conversation patterns and generates prompts

Streaming Transcripts

Sonesse provides a WebSocket endpoint for real-time transcript streaming. Each event contains the speaker, text, and precise timestamps:

import asyncio
import websockets
import json

async def stream_transcript(bot_id, api_key):
    uri = f"wss://api.sonesse.ai/v1/bots/{bot_id}/stream"
    headers = {"Authorization": f"Bearer {api_key}"}

    async with websockets.connect(uri, extra_headers=headers) as ws:
        async for message in ws:
            event = json.loads(message)
            if event["type"] == "transcript.partial":
                yield {
                    "speaker": event["speaker"],
                    "text": event["text"],
                    "timestamp": event["timestamp"],
                    "is_final": event["is_final"]
                }

Building the Coaching Engine

The coaching engine buffers conversation context and triggers analysis when it detects key moments — such as objections, pricing discussions, or competitor mentions:

COACHING_TRIGGERS = [
    "too expensive", "competitor", "not sure",
    "budget", "timeline", "decision maker"
]

async def analyse_for_coaching(transcript_buffer):
    recent_text = " ".join([t["text"] for t in transcript_buffer[-10:]])

    # Check for trigger phrases
    triggered = any(t in recent_text.lower() for t in COACHING_TRIGGERS)

    if triggered:
        coaching_prompt = await generate_coaching_tip(recent_text)
        return coaching_prompt

    return None

Deployment Considerations

When deploying real-time coaching, latency is critical. Keep your LLM calls fast by using streaming responses and caching common coaching patterns. Sonesse's low-latency transcription ensures you're working with the freshest possible data.

Ready to build real-time coaching into your product? Get started with Sonesse.