Skip to content
← All casesReal-time StreamingAI

NZT — real-time dialogue navigator

Real-time Whisper STT → GPT-4o generates contextual hints → WebSocket stream to Android; an FSM drives conversation phases. 200+ tests.

At a glance
Real-time: Whisper STT → GPT-4o → WebSocket

Architect & solo developer

solo · solo

  • Python
  • FastAPI
  • Kotlin
  • Whisper
  • GPT-4o
  • WebSocket

Expand the deep-dive ↓

Problem

What doesn't work

In negotiations and interviews there's no time to think — the hint must be contextual and instant, not after the conversation. Recognition + generation + delivery latency has to be imperceptible.

Solution

Architectural approach

A real-time pipeline: Whisper transcribes speech streaming, GPT-4o generates contextual hints, WebSocket delivers them to Android with minimal latency. An FSM tracks the conversation phase (opening / negotiation / closing) and adapts the prompts to it.

My role & contribution

Architect & solo developer

Designed and built the real-time pipeline: streaming Whisper STT, GPT-4o for contextual hints, WebSocket delivery to an Android client, an FSM of dialogue states, a FastAPI backend and a Kotlin client. 378 backend + 16 frontend + 10 e2e tests.

Ready to discuss?

If you need an architect who builds autonomous AI systems — reach out.

Serbia-based · CET/CEST timezone · EU-aligned working hours · International contracts experience