Applied AI Engineer

Lumbra; New York (NY); On-site; USD 150,000 - 300,000; Full time

Benefits offered by this job

Health, dental, and vision insurance

401(k) with match

HSA plan (optional)

Paid holidays and time off

Office snacks and meals

Job summary

Lumbra is hiring an Applied AI Engineer for on-site work in New York City. You will turn agentic AI models into production features for analysts across the Intelligence Community and the Department of Defense.

You’ll build agents, retrieval, and tool integrations in Nebula, validating them with rigorous evaluation and shipping production code. You will own end-to-end AI features, implement evaluation pipelines, and iteratively improve prompts and tooling based on failure analysis.

Qualifications

• 2+ years shipping production code with strong Python.
• Experience shipping at least one LLM-powered feature to real users.
• Hands-on with retrieval pipelines, tool use, or agent orchestration.
• Experience designing evaluations for non-deterministic outputs.
• Strong software fundamentals: testing, code review, typed interfaces.
• Ability to turn loose asks into working features without a fully specified ticket.
• Strong distributed systems and SOA background.
• Eligibility to obtain a U.S. security clearance.

Responsibilities

• AI features end to end, from prototype to production: agent behaviors, retrieval, tooling.
• Evaluation with datasets, graders, and regression suites.
• Iterate prompts, retrieval, and tooling based on failure analysis.
• Production quality for model calls: latency, cost, error handling, observability.
• Vendor neutrality across model providers, including classified networks.
• Weekly field feedback loop; turn learnings into product improvements.
• Build and scale the agent loop and execution runtime.
• Develop evaluation harnesses and stress-test pipelines.

Skills

Python

Go

Rust

Distributed systems

System design

Tools

Kubernetes

Docker

Cloud platforms (GCP/AWS/AZ)

Job description

About Lumbra

Lumbra builds the vendor-neutral operating system for agentic AI in high-consequence environments. In our first ten months we won four fully competed prime contracts across the U.S. Government, every one of them deploying into classified spaces. We are live on JWICS. We were selected from a field of 200+ companies for a defense innovation challenge and won it. We beat two established defense primes on the same program and earned the follow-on, now in execution.

About the Role

Nebula, our agentic harness, turns days of manual synthesis into minutes of analysis for analysts across the Intelligence Community and the Department of Defense. Whether that is true on a given day comes down to the intelligence layer: how agents reason, retrieve, call tools, and produce output an analyst can trust.

We want an Applied AI Engineer to turn models into product features that analysts actually use. You build agents, retrieval, and tool integrations into Nebula, you prove they work with evaluation rather than demos, and you ship them as production code.

This role is on site in New York City. It is not eligible for remote or hybrid work.

How the Role Is Structured

• ~60% building. Agent behaviors, retrieval pipelines, tool integrations, and the product surfaces around them, shipped to production.
• ~25% evaluating and iterating. Eval sets, graders, regression suites, and failure analysis that tell you whether a change made a feature better.
• ~15% closing the loop. Working with Forward Deployed Engineers and Deployment Strategists to turn feedback from classified deployments into fixes.

What You'll Own

• AI features end to end, from prototype to production: agent behaviors, retrieval, tool use, and the interfaces analysts touch.
• Evaluation for everything you ship. Datasets, graders, and regression suites that catch capability loss without overfitting to specific outputs.
• Evidence-driven iteration on prompts, context, retrieval, and tooling, driven by failure analysis rather than intuition.
• Production quality for model calls: latency, cost, error handling, fallbacks, and observability.
• Vendor neutrality. Features that hold up across model providers, including the models available on classified networks.
• The feedback loop with the field. Field learnings come back weekly, and you turn them into product.
• The agent loop and execution runtime. Build and scale the foundation that sends requests to models, manages tool execution and long-running sessions, and feeds results back into the model's context.
• Evaluation harnesses. Automated benchmarking pipelines and eval suites that stress-test accuracy, latency, context-window limits, and regressions.
• Retrieval and orchestration architecture. Reliable RAG frameworks and orchestration layers that carry high-throughput, low-latency AI features.

You'll Work Alongside

CIA officers, MIT engineers, Special Operations combat veterans, and applied AI engineers from frontier labs including Anthropic and Google. You will make decisions that immediately affect national security. You will own part of a company early and when it matters.

What This Role Is Not

• A research role. You read research to build better features, and you ship.
• A prompt-only role. You write production software, and prompts are one tool among many.
• A demo role. A feature is done when analysts rely on it and you can show that it works.

Required Qualifications

• 2+ years of professional software engineering experience shipping production code, with strong Python. Candidates with 4+ years will be considered at the senior level.
• Experience shipping at least one LLM-powered feature to real users, and knowing how you knew it worked: the eval, the metric, and what you changed because of it.
• Hands-on experience with at least one of retrieval pipelines, tool use and function calling, or agent orchestration.
• Experience designing evaluations for non-deterministic output and iterating on prompts, retrieval, and tooling based on what fails.
• Strong software fundamentals: testing, code review, typed interfaces, and code that survives constant change.
• Comfort taking a loose ask and turning it into a working feature without a fully specified ticket.
• Strong fundamentals in scalable distributed systems, service-oriented architectures, networking, and systems design, ideally having owned both the technical vision and the execution of a foundational platform system end to end.
• Proficiency in Python, Go, Rust, or a similar systems language.
• Experience with cloud infrastructure (GCP, AWS, or Azure) and container orchestration (Kubernetes).
• Full ownership of your work from design through deployment and operations.
• The ability to navigate ambiguity and make sound technical decisions independently, ideally across both 0-to-1 and more mature team or company settings.
• Willingness to work on site in New York City.

Preferred Qualifications

• Experience with evals, agents, or retrieval pipelines at scale.
• Experience with Model Context Protocol or similar tool-use frameworks.
• Experience with open-weight models, self-hosted inference, or constrained compute environments.
• Experience with graph databases or vector search for knowledge representation and retrieval.
• Enough TypeScript and React to build the UI for the features you ship.
• Exposure to intelligence analysis, OSINT, or other analytical workflows.
• Eligibility to obtain a U.S. security clearance.

Compensation

The base salary range for this role is $150,000 to $300,000, spanning our Engineer II and Senior Engineer levels. Where an offer lands depends on scope, autonomy, and technical capability, which we assess through the interview process rather than by years or title. Our ranges are a real guide, and we are flexible for the right person: if your experience clearly exceeds this role's scope, we will consider you at a higher level, with that level's range. Base salary is one part of total compensation, alongside equity and the benefits below.

What We Offer

• Health, dental, and vision insurance. 99% of your premiums covered, and 100% of your dependents' premiums.
• Optional HSA plan. Lower premium, HSA eligible, fully portable, self-directed or managed.
• Health Reimbursement Account for qualified healthcare expenses.
• Life insurance of $150,000 plus AD&D, 100% company paid.
• 401(k) with a 100% match on the first 6%, immediately vested.
• Wellhub Gold. Everything from Gold's Gym to Lifetime, fully covered, plus [solidcore], F45, and hundreds of other studios.
• Unlimited PTO. Real time off, not accrual-based. Take what you need, no counting required.
• Every federal holiday, plus additional paid days off over the winter break.
• Supportive leave of absence program, including time off for military service, medical events, and parental leave.
• Breakfast, lunch, and endless snacks every day you are in the office.
• Pre‑tax commuter benefits.
• Company phone plan. A new iPhone, fully company paid.