Senior Voice Infrastructure Engineer

AethexAI; Indiana (PA); On-site; USD 150,000 - 230,000; Full time

Job summary

AethexAI is building a latency‑aware voice platform and seeks an experienced infrastructure engineer to own GPU‑driven model serving at scale. You will optimize fleets for cost per minute while maintaining strict latency budgets across regions.

Based in London and working on an in‑office team of ~10, you will drive reliability, observability, and fast architectural decisions to keep the pipeline responsive and secure.

Qualifications

• 5+ years running production infrastructure with a focus on latency
• Hands-on GPU inference serving at low latency and GPU fleet cost management
• Experience with Kubernetes, cloud infra (AWS preferred) and infra-as-code

Responsibilities

• Own serving STT, LLM, and TTS model pipelines on GPUs within latency budgets
• Drive cost-per-minute as a core metric
• Ensure reliability, observability, and on-call readiness across the platform
• Maintain security and compliance for a multi-tenant voice system across regions

Skills

Kubernetes

AWS cloud infra

Infra-as-code

GPU inference serving

Cost optimization

Observability

Security & compliance

Multi-tenant systems

Tools

vLLM

Triton

TensorRT-LLM

Job description

The problem

Every voice conversation on our platform is a race against a latency budget: speech in, transcription, a language model, speech back out, all on GPUs, all in the time before a human starts to feel the lag. We run that pipeline across Africa and the Middle East, over noisy lines and real dialects, under infrastructure constraints most companies never have to design around. You'll own the systems that make it fast, reliable, and affordable at scale.

Why it's hard here

GPU is where the difficulty and the money live. You'll be orchestrating model serving across a GPU fleet, keeping it warm enough to hit latency targets and lean enough that cost-per-minute keeps falling. Scale-to-zero pools, warm spares, capacity strategy, and the observability to know what any of it is doing under load. This is the core of the job, not a side quest.

What You'll Own

• Serving STT, LLM, and TTS models on GPU under hard latency budgets, and the fleet behind them
• Cost-per-minute as a first-class metric you're accountable for driving down
• Reliability, observability, and on-call posture across the whole platform
• Security and compliance for a multi-tenant system handling voice and PII across regions
• The major architectural calls, made independently and fast

What We're Looking For

• 5+ years running infrastructure in production, not just building it
• Hands-on GPU inference serving at low latency, and managing GPU fleets and their cost
• Fluency with Kubernetes, cloud infra (AWS ideally), and infra-as-code
• Familiarity with the inference serving ecosystem (vLLM, Triton, TensorRT-LLM or similar)
• A track record of owning complex systems end to end, and shipping from zero with little

Nice to have

• Real-time or low-latency background: telephony, streaming, audio, or voice
• First infra hire or founding-era engineer at an early-stage startup
• FinOps instincts at scale
• Genuine interest in emerging markets or speech tech

The setup

You'll work directly with our CTO on a team of ~10 that ships constantly. High ownership, short path to decisions, no one leaving you alone when something breaks at 2am. London-based, in-office.