Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Build a Realtime Voice Agent

Build a Realtime Voice Agent

Part of InferSights initiative by Simplismart.ai

Avatar for Pratik Parmar

Pratik Parmar

November 09, 2025
Tweet

More Decks by Pratik Parmar

Other Decks in Programming

Transcript

  1. infersights #001 infersights #001 Build Real time Voice Agent with

    STT, LLM and TTS using Simpismart Contact [email protected] Hi, I am here for an appointment I am h Simplismart
  2. Simplismart Hello, I need help with my internet connection. It

    keeps disconnecting. Hi! I’m here to help. Can you tell me if the problem happens on all devices or just one?
  3. Simplismart Our Team - October 2025 Pratik Parmar // DevRel

    linkedin.com/in/pratikparmar1/ Sasmit Datta // ML Engineer linkedin.com/in/sasmit-datta/ Arinjay Saxena // CEO’s Office linkedin.com/in/arinjay-saxena/
  4. Simplismart The Problem - October 2025 According to a report

    by Plivo- 73-79% 12 minutes $4-8 per ticket of customers expect instant support Average wait time: Support costs: BEFORE AFTER
  5. Simplismart Traditional Ways of Support - 1 2 3 4

    5 6 7 8 9 0 IVR Systems Automated telephony system enabling caller interaction via voice or keypad Chatbots AI-powered software that simulates human conversation through text interactions 08:15 AM Assistant Didn’t understand. Try again later. 08:16 AM Help Please... Just Now Assistant Didn’t understand. Try again later. 08:16 AM Custom Development Requires significant time and effort to build a voice agent that convincingly mimics human behavior.
  6. Simplismart Why Pipecat + Simplismart ? October 2025 Pipecat Real-time

    Voice & Multimodal Agent Framework Simplismart MLOps platform offering seamless support for the best models tailored to your specific use case
  7. Simplismart Tech Stack October 2025 STT - Whisper V3 LLM

    - Gemma 3 1B TTS - Kokoro Infrastructure - Simplismart Framework - Pipecat Kokoro Gemma 3 1B Whisper V3 Simplismart TTS, STT & LLM models are hosted on Simplismart Pipecat
  8. Simplismart Voice Agent Architecture October 2025 Voice Input STT LLM

    TTS Voice Output Whisper V3 Gemma 3 1B Kokoro End User Simplismart
  9. Simplismart Deployment through Simplismart October 2025 TTS through docker-hub: LLM

    OpenAI Whisper Deploy Endpoints simplismart/kokoro-tts:latest through Simplismart’s Model Marketplace through Simplismart’s Model Marketplace through Platform Generation
  10. Simplismart Results October 2025 Metrics Before Simplismart After Simplismart First

    Response Time 12 min. < 400ms Resolution Rate 60% 85% Cost per Interaction $ 4-8 $ 0.50 Metrics Before Simplismart After Simplismart First Response Time 12 min. < 400ms Resolution Rate 60% 85% Cost per Interaction $ 4-8 $ 0.50