Voice AI Engineer

NAIRON GmbH; United States; À distance; USD 101 000 - 169 000; Plein temps

Résumé du poste

NAIRON GmbH, a Swiss startup building AI voice agents, seeks a senior/strong-mid engineer located in Greece, Germany or Austria to optimize real-time audio, speech models and latency.

You will work on Python, asyncio, streaming pipelines, LLM APIs, and GPU-based model serving with vLLM, LiveKit, WebRTC. English required; German language skills a plus.

Qualifications

• Experience as a senior or strong mid-level engineer with production experience in real-time audio, conversational AI or ML model serving.
• Strong Python skills, including asyncio and streaming pipelines, with hands-on work on STT/TTS/VAD and latency-sensitive systems.
• Experience with GPUs and model serving tooling (e.g., vLLM) and familiarity with batching and memory constraints.

Responsabilités

• Improve turn-taking behavior and end-of-turn detection in voice agents.
• Reduce end-to-end latency across speech recognition, language models and synthesis; ensure seamless provider failover.
• Optimize use of LLMs in conversation: prompting, tool calls, streaming, parallel requests.
• Serve own speech models (STT and TTS) on GPUs; measure throughput and time to first audio.
• Fine-tune speech and language models for customer languages and voices.
• Build evaluation framework: test calls, benchmarks and quality metrics to detect meaningful improvements.

Connaissances

Python

Asyncio

Streaming

LLM APIs

Outils

vLLM

LiveKit

WebRTC

pipecat

Description du poste

Intro For this position we are only looking for candidates located in Greece, Germany or Austria . Nairon GmbH is a Swiss startup building AI voice agents that answer and make phone calls for businesses. Our platform runs real-time conversations over the phone and on the web, connected to our customers' telephony and business tools. We value clear communication, thoughtful engineering, and a collaborative approach to problem-solving. Join us as we continue to develop tools that make a difference.

Tasks

• Improving turn-taking: when the agent speaks, waits, yields or gets interrupted, including voice activity and end-of-turn detection.
• Cutting end-to-end latency across speech recognition, language models and speech synthesis, and making provider failover seamless.
• Optimising how we use LLMs in conversation: prompting, tool calls, streaming and speculative or parallel requests.
• Serving our own speech models (STT and TTS) on GPUs, measuring throughput and time to first audio.
• Fine-tuning speech and language models for our customers' languages and voices.
• Building evaluation: test calls, benchmarks and quality metrics that tell us when a change really helps.

Requirements

• Experience: A senior or strong mid-level engineer with production experience in real-time audio, conversational AI or ML model serving.
• Technical Skills: Required: Strong Python , including asyncio and streaming pipelines. Hands-on work with speech (STT, TTS, VAD) and LLM APIs in latency-sensitive systems. Model serving on GPUs, for example with vLLM or similar, and the basics of batching and memory limits.
• Nice to have: LiveKit Agents , pipecat or WebRTC. Fine-tuning (LoRA or full) of speech or language models. Audio signal processing such as echo cancellation or noise suppression. German, or experience with German-language speech data.
• Communication: English.