AI Engineer

AI Engineer, Agentic Systems and Multi-Agent Orchestration Location: London (HQ), Hong Kong, Shanghai, UAE, New York About the Role A private investment firm with offices across the globe (London, Hong Kong, Shanghai, Abu Dhabi, and New York) is expanding its elite technical team to build and own state-of-the-art agentic AI infrastructure. The role can be based in any of these locations. You will architect the core systems that let autonomous AI agents plan, execute tool calls, retry on failure, and reliably hand off complex workflows in production. At the centre of this is agent harness engineering: the scaffolding around the model, covering tools, memory, guardrails, and feedback loops, that turns a raw LLM into an agent you can trust in production. You will be working on exciting projects within the quantitative finance world, such as research agents that help generate and test trading ideas, AI tools that process market data, news, and alternative data, and agentic systems that support portfolio and risk monitoring. Your work will feed directly into real trading and investment decisions. Key Responsibilities

• Design and build scalable multi-agent orchestration systems, covering task decomposition, inter-agent communication, and state management
• Own agent harness engineering, including sandboxed execution, guardrails, automated retries, fallback logic, and real-time observability
• Build agent memory and context engineering layers that keep long-running agents coherent across sessions and tool calls
• Build rigorous LLM evaluation pipelines, including LLM-as-judge and trajectory-level testing, to catch regressions before production deployment
• Architect reliable RAG and tool-use infrastructure across structured and unstructured data, using hybrid search and reranking, with MCP and agent-to-agent protocols for integration
• Fine-tune and adapt models for domain-specific tasks, and contribute to model pretraining and post-training work where it adds an edge
• Own ML deployment end to end, taking models and agents from experiment to production-grade serving, including monitoring, versioning, and cost and latency management
• Work across frontier commercial models and open-source models, routing between them based on cost, latency, and task type What You Bring
• Deep expertise building with advanced LLM architectures, agentic AI, agentic reasoning, and multi-agent workflows
• Hands-on experience with agent harnesses and orchestration, using LangGraph, AutoGen, or custom-built frameworks, alongside Model Context Protocol
• Practical experience with model fine-tuning, and ideally exposure to pretraining at meaningful scale
• Experience deploying ML models and LLM-based systems into production, with strong MLOps fundamentals
• Strong grounding in LLM evaluation, including designing benchmarks and metrics that reflect real-world performance
• Experience with RAG and vector databases such as Pinecone, Qdrant, or Chroma, and hybrid retrieval pipelines
• Strong Python, including AsyncIO and FastAPI, with PyTorch or TensorFlow for model-level work
• A clear grasp of where standard agent approaches such as ReAct loops break down, and how to engineer resilient workarounds
• Background in financial services, investment management, fintech, or high-throughput engineering environments is advantageous
• Open-source contributions, published AI research, or relevant patents are highly valued Compensation and Benefits
• Highly competitive compensation package
• Discretionary performance-based annual bonus
• Full visa sponsorship and premium relocation package where applicable
• Comprehensive private health insurance and dependent schooling allowances How to Apply If you want to build cutting-edge agentic infrastructure in a high-performance environment, apply now or reach out directly to Syed at syed@durlstonpartners.com.