Lumbra is building Nebula, an agentic harness that makes AI agents reliable, evaluable, and useful for real analytical work in high-consequence environments. We're looking for an Applied AI Engineer to own the intelligence layer of the harness: how agents reason, call tools, retrieve knowledge, and produce trustworthy outputs. You'll build production systems and research prototypes in equal measure, working at the boundary of what's possible with artificial intelligence today.
What You'll OwnDesign and evolve our provider-agnostic LLM integration layer, maintaining a unified abstraction over streaming, tool calling, structured output, and context management that lets the platform run multi-turn, multi-agent conversations across models with no configuration changes.
Develop advanced agent orchestration techniques that combine deterministic and non-deterministic tool use to produce complex, multi-step action chains. Outputs must be traceable, interpretable, and grounded in robust, verifiable citations.
Build MCP servers over internal data sources and systems, designing agent-native interfaces that go beyond API wrappers to reduce ambiguity, improve determinism, and give agents first-class access to organizational knowledge.
Extend our MCP-based tool interaction framework with capabilities like structured citations, multimodal tool output, and rich provenance metadata that let agents reason transparently about the results they consume.
Architect the agentic memory layer for short-, medium-, and long-term recall across tasks and projects, spanning vector, text, graph, and filesystem stores. Design retrieval strategies that efficiently surface the right context to agents at scale.
Experience building production agentic systems, including the infrastructure for reliability, observability, and evaluation, not just prompt engineering
Background in evaluation methodology for AI systems: offline benchmarks, human-in-the-loop review, inter-annotator agreement, or similar quality measurement frameworks
Familiarity with prompt engineering at scale: template management, few-shot optimization, chain-of-thought patterns, and systematic prompt testing
Prior work with the Model Context Protocol (MCP) or similar tool-use standards for agent-to-system interaction
Experience with information retrieval systems: search ranking, relevance scoring, or hybrid keyword and semantic search
Understanding of AI safety practices: output filtering, hallucination detection, constrained generation, or grounding techniques
Prior work in analytical domains (intelligence, investigations, due diligence, research, competitive intelligence, or similar) that informs what "useful AI" means for practitioners making high-stakes decisions
BenefitsComprehensive medical, dental, and vision plans
Premiums 100% covered by Lumbra for all employees
Exceptionally low premiums for spouses and dependents
Basic life insurance and disability 100% covered for all employees by Lumbra
Option to purchase additional life insurance available
Take the time off that you need, when you need it' paid time off, not accrual based
Generous company holiday calendar including a holiday shutdown in December
Supportive leave of absence program including time off for military service, medical events, and parental leave
Full 401(k) retirement plan for all full-time eligible employees
Company-funded commuter benefits
Free access to on-site gym at office
Skills Required
- Design and maintain a provider-agnostic LLM integration layer (streaming, tool calling, structured output, context management).
- Develop advanced agent orchestration techniques combining deterministic and non-deterministic tool use with traceability and verifiable citations.
- Build MCP servers over internal data sources and design agent-native interfaces to organizational knowledge.
- Extend MCP-based tool interaction frameworks with structured citations, multimodal tool output, and provenance metadata.
- Architect agentic memory layer spanning vector, text, graph, and filesystem stores and design scalable retrieval strategies.
- Experience building production agentic systems including infrastructure for reliability, observability, and evaluation.
- Background in evaluation methodology for AI systems (offline benchmarks, human-in-the-loop review, inter-annotator agreement).
- Familiarity with prompt engineering at scale: template management, few-shot optimization, chain-of-thought patterns, systematic prompt testing.
- Prior work with the Model Context Protocol (MCP) or similar tool-use standards.
- Experience with information retrieval systems: search ranking, relevance scoring, hybrid keyword and semantic search.
- Understanding of AI safety practices: output filtering, hallucination detection, constrained generation, grounding techniques.
- Prior experience in analytical domains (intelligence, investigations, due diligence, competitive intelligence, or similar).
What We Do
Lumbra is an AI company building the architecture for autonomous intelligence in the intelligence community. They are developing frameworks, orchestration layers, and agentic operating systems, including Nebula—an agentic harness designed to make AI agents reliable, evaluable, and useful for real analytical work in high-consequence environments. The team consists of engineers and operators from the intelligence community, special operations, and frontier AI research.
