- $180k-$220k + super + bonus

Most “AI engineer” roles mean wiring an LLM into a chatbot. This one is about the hard part: making agents that take real, consequential actions and get them right every time.

Our client is building an agent-driven travel platform. Its agents turn natural-language requests into live bookings and servicing work across supplier APIs, reservation terminals, company travel policies and persistent booking records. When a supplier times out halfway through a booking, the system has to know what actually happened before it tries again.

They’re looking for a senior engineer to own the engineering around the model, working directly with the founder and owning much of the architecture.

What you’ll work on

- Agent runtime and harness design: orchestration, checkpointing, cancellation, budgets, concurrency and handling provider failures
- Context engineering and persistent memory, including history compaction, prompt-cache efficiency and memory scoped correctly to each user and company
- Typed tools and MCP integrations, keeping model intent separate from deterministic validation and execution, with protection against prompt injection
- Reliable transactions: idempotency, reconciliation, authoritative readback and approvals tied to the exact action being approved
- Subagent delegation and longer-running workflows
- Evaluation and observability: replay suites,



correlated traces, and tracking task success, latency and cost per task

The stackWhat you’ll bring

- Robust production Python, including asyncio, typed interfaces and debugging concurrent systems
- You’ve shipped stateful, tool-using LLM agents to real users with external integrations
- You can explain how your agent framework or custom runtime actually executes, not just how to call it
- Solid backend depth: Postgres, transactions, migrations, caching and distributed-system failure modes
- You’ve built evals for probabilistic systems and used traces to diagnose failures
- Security fundamentals: tenant isolation, least-privilege tool access and handling sensitive data

Bonus:

- contributions to agent SDKs, MCP servers or eval frameworks
- GDS/NDC or booking/payment workflows
- terminal-based or multi-channel (web, messaging, voice) agents
- model routing and inference optimisation

Why it’s worth a conversation

You’ll have real ownership from day one. In your first 90 days you’ll map the runtime, set a quality, latency and cost baseline, and ship a harness improvement you can measure.

If you’ve built an agent that did meaningful work through tools, and you can talk through how you diagnosed a failure and proved the fix worked, we'd love to hear from you.

📌 Artificial Intelligence Engineer (New South Wales)
🏢 Talent Forge Group
📍 New South Wales