AI System Engineer

Focus: MCP, Agentic AI, Tool-Oriented LLM Architectures

We are engaging senior AI contractors to design and deploy production-grade AI systems for one of our clients - a global leading strategic management consultancy for their technology build unit.

This is not a chatbot build role. This is systems architecture around LLM-driven agents operating across real tools, real data, and real constraints.

• Design and implement Model Context Protocol (MCP)-style tool ecosystems
• Build secure tool servers with structured schemas (JSON-based tool definitions)
• Architect multi-step agent workflows (planning, execution, reflection loops)
• Implement persistent memory and retrieval (RAG, embeddings, vector DBs)
• Design guardrails and deterministic fallback logic
• Build observability into LLM pipelines (cost, latency, failure analysis)
• Deploy production-ready systems (cloud-native, containerized)

You will work directly with client core consultant team and to move from prototype to production.

• Strong understanding of transformer-based models
• Experience with OpenAI / Anthropic APIs
• Embedding pipelines and RAG architecture
• Context window management and structured prompting
• Model evaluation beyond it works

Familiarity with ecosystems such as Open AI, Claude etc.

• ReAct / multi-step reasoning architectures
• Tool arbitration logic
• Multi-agent coordination
• State persistence across sessions
• Reflection/self-correction loops

Framework exposure helpful but not sufficient e.g. Lang Chain, Microsoft Auto Gen, CrewAI. You must be able to build beyond frameworks.

• Python (FastAPI preferred) or equivalent backend stack
• Vector databases (Pinecone / Weaviate / Milvus or similar)
• Containerization (Docker)
• Cloud deployment (AWS / Azure / GCP)
• Logging, telemetry, and cost optimisation

If you have not deployed at scale, this may not be the right engagement.

• Prompt injection mitigation
• Tool permission boundaries
• Data isolation strategies
• Audit logging
• Safe failure modes

Our clients operate in IP-sensitive and regulated environments.

• Fine-tuning experience (LoRA, PEFT)
• On-prem or private model deployment
• Multi-agent simulation environments
• Hybrid symbolic + neural systems
• Experience building internal AI platforms rather than one-off tools

• Prompt engineers without backend capability
• Demo builders without production exposure
• Pure research profiles with no delivery experience

We value engineers who think in systems, not scripts. Looking forward to hearing from you.