Agentic AI Engineer

Cloud People; Greater London; Hybrid; GBP 110,000 - 190,000; Full time

Job summary

Cloud People is building a production-grade agentic AI pod from London with autonomy and strong governance. The role focuses on turning LLM concepts into reliable services that can integrate into enterprise banking environments, with minimal runtime disruption and robust failure handling.

You will design, build and operate production AI systems, owning patterns for agents, memory, retrieval and observability while ensuring cost-effective performance across Azure OpenAI, Anthropic and open models.

Qualifications

• 4+ years in software, machine learning or production AI engineering.
• 2+ years with LLMs, prompt/context engineering and production monitoring.
• Hands-on experience in agent architecture and orchestration.
• Experience with LangGraph, LangChain, LlamaIndex, Semantic Kernel, AutoGen or CrewAI.
• Strong Python engineering including async, type hints, API design and tests.

Responsibilities

• Design, build and operate production agentic AI systems that reason, use tools, maintain state, retrieve knowledge and escalate safely to humans.
• Own agent architecture, orchestration and decision loops, including multi agent patterns.
• Select and integrate LLMs across Azure OpenAI, Anthropic and open models while balancing latency and cost.
• Engineer reliable tool use with function calling, structured outputs and audit trails.
• Deploy agents as reliable services using CI/CD and proper environment separation.

Skills

Python

LLMs

Async programming

Distributed systems

Observability

Security & governance

Tools

LangGraph

LangChain

LlamaIndex

Semantic Kernel

AutoGen

CrewAI

Job description

London based, one office day per week, with some travel to the UAE.

Company & role

This role sits with a global IT solutions provider standing up something genuinely different: an autonomous rapid prototyping pod for banking, insurance and fintech clients. A small, elite team that wins its own work, takes an ambiguous client problem, and turns it into a working AI prototype in four to five weeks, then hardens it into production. No layers, no handoffs, no waterfall. Just building.

This is the technical brain of that pod. You will design, build and operate production agentic AI systems that reason, use tools, hold state, retrieve knowledge, run multi step workflows and escalate safely to a human when they should. The focus is turning LLM concepts into reliable, observable services with proper failure handling and guardrails, and making them integrate cleanly into enterprise banking environments, because a solution that can't integrate is a dead solution.

This is a deeply technical role, not a client facing one. You'll be left alone to build.

Why This Role Stands Out

The pod is deliberately built for engineers who want to engineer. It is fully autonomous and protects your time: no L2 tickets, no site reliability duties, no being pulled onto other teams' problems, no meeting culture. You design and build agents, and that's it.

The work itself is production agentic AI in a setting where reliability, safety and cost have genuine consequences, with real depth across architecture, orchestration, retrieval, evaluation and observability. You'll help set the patterns for how agents are built and governed here, with a modern stack and GPU inference behind you, and mostly remote working with one office day a week.

There is a growing UAE dimension to the business too. Nothing is expected, but if working in or relocating to the UAE would ever appeal, they will back you to do it.

Key Responsibilities

• Design, build and operate production agentic AI systems that reason, use tools, maintain state, retrieve knowledge and escalate safely to humans
• Own agent architecture, orchestration and decision loops, including multi agent patterns, memory design and multi turn conversation handling
• Select and integrate LLMs and agent frameworks across Azure OpenAI, Anthropic and open models, balancing latency, cost, quality, data residency and compliance
• Engineer reliable tool use, including function calling, structured outputs, API wrappers, permissions, retries, timeouts, sandboxing and audit trails
• Implement retrieval, memory and context patterns including RAG, hybrid search, re ranking, summarisation and context budgeting
• Mitigate hallucination and handle failure properly, with clear recovery and escalation paths
• Integrate agents into enterprise systems and client tooling to banking grade requirements
• Own agent evaluation, safety and observability, including automated evals, golden datasets, red team testing, prompt injection defences, PII controls, tracing and dashboards
• Optimise performance and commercial viability through token budgeting, prompt caching, model routing, batching and cost monitoring
• Deploy agents as reliable services using CI and CD, environment separation, secrets management, feature flags, canary releases and rollback

Ideal Experience

• 4 or more years in software, machine learning or production AI engineering, with evidence of shipping reliable services beyond prototypes
• 2 or more years working with LLMs, across prompt and context engineering, structured outputs, function calling, RAG, evaluation and production monitoring
• Genuine, hands on experience of agent architecture and orchestration you have personally designed or built, not framework assembly
• Experience with frameworks such as LangGraph, LangChain, LlamaIndex, Semantic Kernel, AutoGen or CrewAI, or comparable custom orchestration
• Strong Python engineering, including async programming, type hints, validation, API design, test automation and secure, maintainable service architecture
• A real understanding of enterprise integration, data layers, hallucination mitigation and failure handling
• Security and governance knowledge covering prompt injection, data exfiltration, PII handling, sandboxing, approvals and audit evidence
• Financial services, banking, insurance or fintech experience, or other regulated environments
• Agent evaluation tooling such as LangSmith or Braintrust, and observability with Application Insights
• Multimodal agents, and Arabic or UAE localisation
• On premise or local inference with vLLM or TensorRT LLM, and model routing for cost and performance