AI Lead

Job Title: AI Lead

Location: Brookvale, New South Wales 2100

Work Model: On-Site

Overview:

Red Cartel is looking for an AI Lead to own our applied AI platform.

This is not a role about calling a hosted model API. It is a real-time system built and operated end to end in house, running on self-managed infrastructure, where latency, reliability and the behaviour of non-deterministic models are all first-order engineering problems.

We want an engineer who has built something of this shape and now wants to lead it. You will stay hands-on in the code while setting technical direction.

Core Responsibilities:

- Own the platform end to end: backend services, orchestration, inference, internal tooling and client SDKs.

- Own a real-time voice pipeline: voice activity detection, streaming speech recognition, turn taking and interruption handling, response generation and speech synthesis, inside a latency budget tight enough that the interaction feels immediate.

- Own how model behaviour is specified and evaluated: how an intended behaviour is authored, and how it becomes a repeatable outcome rather than a one-off result.

- Own the evaluation loop: the automated systems that prove changes to prompts and model configuration made things better rather than worse, without relying on manual review.

- Decide where inference runs. Balance self-hosted against hosted models per workload on cost, latency, privacy and customer data handling constraints.

- Own deployment of GPU inference across mixed hardware, containerised services, migrations and the production upgrade path.

- Raise applied AI capability across the wider engineering group.

What We're Looking For:

Systems and Backend:

- Strong systems programming background with production Go. Deep experience in another systems language plus clear evidence you pick up Go fast will be considered.

- Proven experience with concurrent, event-driven or pipeline architectures where ordering, back-pressure and race conditions are real correctness problems.

- Strong Python, including the machine learning tooling around it.

- Solid PostgreSQL and data modelling, including migrations against live production data.

- Redis, stream processing or message brokering, and background job orchestration.

AI and Machine Learning:

We are especially interested in candidates who can speak concretely to their experience in some of the following:

- Running large language models in production, including prompt architecture, structured output, context management and cost control.

- Self-hosting inference: running and tuning LLM, speech recognition and text-to-speech models on your own GPU hardware.

- Real-time speech pipelines: voice activity detection, streaming ASR, forced alignment, TTS and voice cloning, and the latency engineering that makes them usable in conversation.

- Automated evaluation of non-deterministic systems: model-as-judge, simulated users, and detecting regressions in model behaviour.

- Fine-tuning, including LoRA and parameter-efficient methods, and the judgement to know when fine-tuning is the wrong answer.

- Model quantisation and the practical trade-offs of running inference on constrained or varied hardware.

Real-Time Media and Infrastructure:

- WebRTC or a comparable real-time media stack: transport, session management, mixing and recording.

- Audio fundamentals: sample rates, resampling, buffering, codecs and the failure modes of streamed audio.

- Strong Docker, including programmatic container lifecycle management and GPU-enabled workloads.

- CI/CD design with meaningful automated gates, and honest handling of tests that need hardware CI cannot provide.

Client-Facing Skills:

- Working TypeScript and React, enough to own an internal tooling frontend and review others' work on it.

- C# and enough Unity package familiarity to own a client SDK that our engineering teams consume.

Leadership:

- Demonstrated technical leadership of a team or a significant platform.

- Ability to make sound trade-offs under delivery pressure and to say no with a defensible reason.

- Documentation and specification discipline. We write subsystem specs and record the reasoning behind decisions.

Preferred Qualities:

- Experience delivering a real migration or rewrite of a live system without losing the users on it.

- Comfort working with defence and law enforcement customers, and careful reasoning about data handling, privacy and on-premise constraints.

- Eligibility to hold, or to obtain, an Australian security clearance.

- Deep understanding of the AI product ecosystem: what is emerging, what is stable, and what is hype.

- Experience on early-stage or fast-moving product teams.

Who are we?

Red Cartel is a wholly-owned subsidiary company of xReality Group Limited (XRG). XRG acquired Red Cartel in August 2021 to enhance its in-house development capabilities.

At XRG, we're pioneering the frontier of immersive XR experiences for elite military and law-enforcement teams. Our flagship offering, Operator XR, is transforming tactical training by combining portability, realism and security. Housed in a single 18 kg case, it allows agencies to train anywhere, anytime using a combination of their own equipment and our custom peripherals. Operator XR provides instructors with fully customisable scenarios and deep analytics that are accessed via our custom After Action Review (AAR) component.