Lead AI System Architect

Lead AI System Architect

Department: 501 Product and Technology

Employment Type: Full Time

Location: Remote, United Kingdom

The AI System Architect leads the architecture of EIS's agentic AI platform — the design of multi-agent systems that automate insurance workflows end-to-end across Policy, Billing, and Claims domains. The role owns the patterns, frameworks, and standards for agent orchestration, MCP-based tool ecosystems, agent memory, planning, evaluation, and safety. RAG and conversational features are table stakes; the forward agenda is autonomous and semi-autonomous agents that act on behalf of users - quote intake, claims triage, underwriting and pricing intelligence, billing troubleshooting, and beyond - across our platform and technology stacks.

• Own the architecture of EIS's agentic platform: agent orchestration, MCP-native tool ecosystems, agent memory (short-term, long-term, semantic), planning, and tool/function calling patterns reusable across product domains.
• Enable and provide support for domain teams for vertical insurance agents and the horizontal capabilities (RAG, retrieval, instructional flows) they compose from.
• Define and enforce levels of autonomy — assistive, semi-autonomous, autonomous — with explicit human-in-the-loop checkpoints, escalation paths, and reversibility for high-stakes actions in regulated workflows.
• Drive the MCP strategy: which capabilities EIS exposes as MCP servers to internal and partner agents, how our agents consume external MCP tools, and the tool registry, schemas, and versioning that keep this scalable.
• Maintain the multiple stack approach as a first-class capability:
  Typescript, and Java. Help teams to pick the right stack per agent and keep all aligned through shared configuration artefacts, prompt management, and evaluation tooling.
• Lead Architecture Decision Records (ADRs) for agentic capabilities; partner with Platform, Security/Info Sec, and Dev Ops so agents are observable, testable, sandboxed, and compliant by default.
• Drive AI Dev Ops for agents: trace capture and replay, eval harnesses (task success, tool-use correctness, regression), prompt and model versioning, cost and latency budgets per agent, and progressive rollout strategies.
• Set safe-AI standards for agentic systems: prompt injection and tool-poisoning defenses, action allow-lists, blast-radius controls, PII handling, data residency, and bias mitigation. Treat agent safety as a first-class architectural concern.
• Translate insurance use cases into production agent designs with product strategists and domain architects; provide technical leadership and mentorship; communicate agentic trade‑offs (autonomy, reliability, cost, safety) clearly to executives, customers, and engineers.

• Proven track record designing and shipping agentic systems in production - not demos, not prototypes - with meaningful autonomy and multi-step tool use.
• Strong systems background: data‑intensive, distributed, and latency‑sensitive design in production environments.
• Deep, hands‑on experience with agent patterns: orchestration, planning, ReAct‑style and graph‑based agents, agent memory, tool/function calling, MCP, structured outputs. Sharp instinct for when an agent is the right answer and when to use a deterministic workflow. Tracks the frontier and translates what matters into the roadmap.
• Strong with the Java/Spring ecosystem.
• Strong with Typescript and Python for AI (Lang Chain, Lang Graph, or equivalent agent framework) - production experience required. Equally comfortable in both stacks.
• Hands‑on with vector databases including embedding models, hybrid search, re‑ranking, and retrieval evaluation.
• Experience with agent evaluation and observability: traces, replays, eval harnesses, guardrails, and cost/latency telemetry. Familiar with AI configuration‑as‑code.
• Experience shipping AI services on cloud platforms (AWS, Azure, GCP) in regulated enterprise environments - security review, data residency, audit trails.
• Familiarity with insurance, financial services, or another regulated domain is a plus.
• Strong architectural judgment — pragmatic about build vs.…