click to skip

Back to Portfolio
OnlineModel: gpt-4o-custom

// Case study · Voice AI agent

AI Voice Terminal (Peter)

  1. STT
  2. NLP
  3. Tool call
  4. TTS

01 · Speech-to-text

Input stream

whisper-large-v3 · 16kHz

User

listening…

Peter

—

Peter

Awaiting wake word

loop 1 · 0.0s / 12s

03 · Reasoning

Intent & function calling

tool_choice=auto

› waiting for input

Intent confidence—

02 · NLP

Tokenization & embedding

0/8 tokens · dim 3072
  • Check0x8F35
  • the0x3997
  • camper0x32E3
  • van0x3623
  • batt0x6F60
  • ery0xFDF8
  • levels0xB714
  • .0x40D2

Embedding vector (first 64 of 3072 dims)

Session

Recent calls

First interaction in progress…

Simulated session for demonstration · the real agent runs as a Python service with Whisper STT and OpenAI function calling.