Senior AI/ML Engineer (Agentic foundation) - Fulltime - Worldwide

Turn the agentic foundation from architecture into decisions and running code. You work directly with software engineering team and myself (CPO / cofounder), who owns the agentic system until this hire lands.

What you'll do

• Ratify the agentic foundation plan as architecture decision records: runtime, sandboxing, data stores, scheduling, how agents use our data and clients data and tools, and how the foundation plugs into our applciation infra.
• Build in Python (Google ADK or equivalent), calling Loci tools through the gateway (MCP), with approval before any write
• Design and implement reusable agent runtimes: versioned agent definitions, persistent agent instances, model selection, isolated compute (microVMs / gVisor-class sandboxing or equivalent)
• Stand up retrieval and RAG over Loci (and customer context where appropriate): grounding, evaluation harnesses, and measurable quality for market and portfolio questions
• Ship a first specialist agent over A2A; define evaluation suites, tracing, and model/cost choices so production behaviour is observable and improvable
• Interface with our application ifra Core alongside Alex (Lead Full Stack / Roam OS technical owner, starting 9 Nov) so agents, tools, and host services share a clear contract
• Deploy agent paths for clients and leave handover so Roam can operate and extend the foundation without you

Must-haves

• Senior LLM application engineering in production — agents, tools, or RAG systems that other people depend on
• Strong Python; experience with an agent framework (Google ADK, LangGraph, or similar) and tool protocols (MCP at minimum)
• Production judgement on evals, observability, and model selection — you measure behaviour, not only demo it
• Ability to turn an architecture plan into decision records and a staged build that ships through pull requests
• Comfort with retrieval systems and with exposing agents behind APIs / gateways

Nice to have: A2A or other agent-to-agent protocols, agent sandboxing (microVMs, gVisor, or Kubernetes-based platforms), TypeScript for application Core, Kubernetes basics, enterprise identity and tenant-aware tool access, real-estate or structured market-data experience