Lead AI Engineer, Multi-Agent Systems

Durlston Partners LLP; Abu Dhabi; On-site; AED 300,000 - 600,000; Full time

Benefits offered by this job

Full visa sponsorship

Relocation package

Private health insurance

Dependent schooling allowances

Job summary

Durlston Partners LLP in Abu Dhabi seeks an AI Engineer – Multi-Agent Systems to architect core autonomous AI agent infrastructure, enabling planning, tool calls, retry logic, and reliable handoffs of complex workflows in production.

You will design scalable systems for task decomposition, inter-agent communication, and state management; own sandboxing and guardrails, automated retries, and real-time observability; and build evaluation pipelines for performance before deployment.

Qualifications

• Expertise with LLM architectures, RAG, and multi-agent workflows.
• Understanding where ReAct-like loops break down and resilient workarounds.
• Experience in financial services, fintech, or high-throughput engineering environments is advantageous.
• Active open-source contributions or AI research publications are valued.

Responsibilities

• Design and build scalable systems for task decomposition and inter-agent communication.
• Own sandboxing, guardrails, automated retries, fallback logic, and real-time observability.
• Build rigorous evaluation pipelines to catch performance regressions before production deployment.
• Architect reliable retrieval (RAG) and tool-use infrastructure across structured and unstructured data sources.

Skills

LLM architectures

RAG

agentic reasoning

multi-agent workflows

Tools

LangGraph

AutoGen

Python

PyTorch

TensorFlow

AsyncIO

FastAPI

Job description

Durlston Partners LLP in Abu Dhabi seeks an AI Engineer – Multi-Agent Systems to architect core autonomous AI agent infrastructure, enabling planning, tool calls, retry logic, and reliable handoffs of complex workflows in production.

You will design scalable systems for task decomposition, inter-agent communication, and state management; own sandboxing and guardrails, automated retries, and real-time observability; and build evaluation pipelines for performance before deployment.