AI Architect [45601]

Marionete is an entrepreneurial technology consultancy helping organisations use data, software engineering, cloud platforms, and AI to solve complex business problems.



For more information, visit: www.marionete.co.uk

Lead the architecture, design, and delivery of production-grade agentic AI and Generative AI systems that improve underwriting profitability, operational efficiency, and risk selection across the London Insurance Market.



Combine multi-agent orchestration, LLM productionisation, and Responsible AI with robust engineering to create safe, scalable, repeatable solutions.

Agentic AI & GenAI Architecture Leadership

• Architect end-to-end multi-agent systems: planner/executor and supervisor/worker patterns, tool-calling, memory and state management, and human-in-the-loop checkpoints for high-stakes decisions.
• Design production-grade RAG architectures: chunking and indexing strategies, hybrid retrieval, re-ranking, grounding, and citation/traceability for auditability.
• Define agent evaluation and guardrail frameworks: task success metrics, hallucination and drift detection, cost-per-task, latency budgets, and automated regression testing for prompts and agent behaviour.
• Architect for scale and governance: multi-tenant agent platforms, versioned prompt/tool registries, structured output contracts, and safe fallback/escalation paths.
• Create reusable AI components: RAG pipelines, agent orchestration frameworks, evaluation harnesses, prompt/tool registries, cost-aware inference layers.

LLMOps, Responsible AI & Productionisation



• Build reliable CI/CD and MLOps/LLMOps pipelines using Databricks, MLflow, Unity Catalog, Azure ML, and model/prompt versioning frameworks.
• Implement observability for hallucination rate, drift, accuracy degradation, cost, latency, and compliance risk signals.
• Lead Responsible AI compliance programmes (bias, explainability, audit trail) suited to regulated financial-services environments.
• Apply AIOps principles for automated recovery, adaptive retraining, secure two-gate deployment, and managed online scoring/inference endpoints.

AI & Data Strategy for Insurance Organisations

• Assess agentic AI maturity, data readiness, platform gaps, and engineering risks.
• Produce capability maps, target architectures, TOMs, and AI roadmaps aligned to underwriting and claims profitability levers.
• Advise senior stakeholders on where agentic AI and GenAI can directly impact.

Technical Skills

• 8–15+ years delivering AI/ML systems in production, including 3+ years focused on agentic AI / GenAI architecture at enterprise scale.
• Proven track record shipping multi-agent platforms and LLM-based copilots into production, not just proof-of-concept, ideally across financial services or other regulated industries.
• Hands-on experience with agentic orchestration frameworks (LangChain, LangGraph, Semantic Kernel, AutoGen, CrewAI or equivalent), RAG architectures, and vector search.
• Strong engineering fundamentals: Python, distributed systems, API design (REST/gRPC), cloud-native architectures.
• Working knowledge of classical data-science modelling (regression, time-series forecasting, gradient boosting, clustering, anomaly detection, optimisation) sufficient to integrate agentic systems with existing ML pipelines.
• Experience building feature pipelines, evaluation frameworks, monitoring dashboards, and cost-aware inference layers.

Leadership & Communication

• Led cross-functional teams (engineers, data scientists, PMs) delivering complex AI platforms.
• Confident working with CIOs, Heads of Data, Underwriting Directors, and Claims leadership.
• Demonstrated ability to turn R&D and agentic prototypes into production systems that deliver commercial value (e.g. measurable cost reduction, process agentification, latency improvement).



Domain (Lloyd's of London) (Advantageous)

• Understanding of underwriting workflows, exposure models, pricing models, bordereaux, actuarial inputs, claims triage.
• Ability to design agentic systems that directly support Lloyd's profitability framework, for example:
• Underwriting decision support and pricing segmentation
• Claims triage and claims leakage reduction
• Exposure analysis and cyber risk scoring
• Delegated authority (DA) automation and bordereaux processing
• Document intelligence and operating expense reduction

Agentic AI & LLM

• LangChain / LangGraph / Semantic Kernel / AutoGen / CrewAI (agentic orchestration)
• Vector stores (Pinecone, Weaviate, Milvus, Azure AI Search / Vector DB)
• Evaluation frameworks (LangSmith, deep-eval, Ragas)
• LLM integration (OpenAI API, Azure OpenAI, Anthropic Claude, open-source LLMs)
• Real-time conversational AI / voice assistants — advantageous

Model Lifecycle & Governance

• MLflow
• Databricks Model Serving
• Databricks Unity Catalog
• Feature Store (Databricks or Feast)
• Responsible AI / model monitoring platforms

Data Engineering & Pipelines

• Databricks Lakehouse
• Apache Spark (batch + structured streaming)
• Apache Kafka / Confluent Cloud (event-driven data & AI triggers)
• Delta Lake
• Airflow / Databricks Workflows (orchestration)

Cloud Platforms (at least one)

• Azure (Azure ML, AKS, ADF)
• AWS (SageMaker, ECS, Lambda)
• GCP (GKE, Vertex AI)

Data Science & Modelling (Advantageous)

• Python (Pandas, NumPy, SciPy, Statsmodels)
• Scikit-learn
• XGBoost / LightGBM / CatBoost
• Time-series forecasting frameworks (Prophet, SARIMAX, ML-based forecasting)
• Optimisation libraries (OR-Tools, Pyomo)
• Anomaly detection & segmentation methods



Software Engineering/Productionisation

• Docker
• Kubernetes (AKS/EKS/GKE)
• REST & gRPC APIs
• CI/CD: GitHub Actions, Azure DevOps, GitLab CI
• IaC: Terraform
• Observability: Prometheus, Grafana, ELK

• Competitive salary package;
• Hybrid/remote position;
• Health insurance and life insurance;
• Financial support for personal development and training;
• Relaxed dress code at the Marionete offices;
• An abundance of career paths and opportunities in which to advance;
• A friendly, supportive, flexible and hybrid work environment.

Marionete believes in providing equal employment opportunities to everyone. We do not practice and will not tolerate discrimination on the basis of race, skin color, ethnicity, national origin, gender, sexual orientation, marital status, maternity, religion, age, disability, gender identity, results of genetic testing, or service in the military.

By applying to Marionete vacancies, you agree that your data and CV remain secure and confidential with your application. Marionete is a company in compliance with General Data Protection Regulation (GDPR).