Software Engineer - Applied AI

luminary; Deutschland; Vor Ort; EUR 90.000 - 120.000; Vollzeit

Zusammenfassung

Luminary, the AI-native system for wealth transfer services, seeks an experienced software engineer focused on AI-powered document understanding and knowledge graphs. You will ship production systems and contribute to prompt/context engineering and robust evals.

Join a team that ships frequently, collaborates with product and design, and values reliable, scalable AI features for private clients and firms. Remote-friendly culture and strong engineering practices.

Qualifikationen

• Four or more years of software engineering experience at companies with strong engineering practices.
• An engineer who works on AI, not a researcher.
• Enjoy shipping to production systems.
• Experience with prompt and context engineering, eval design, and making probabilistic systems dependable.

Aufgaben

• Improve accuracy on core extractions and handle long or amended documents with unclear language.
• Own our evals. design and maintain suites that tell us whether changes helped and guard against regressions.
• Improve our shared agent harness: tooling, parallel tool calls, context strategy and orchestration of agentic workflows.
• Design for human review: collaborate with product and design on how suggestions are surfaced, explained, and corrected.
• Be informed on the state of AI: track models and best practices to keep us at the frontier.

Kenntnisse

Software engineering

AI systems

Prompt engineering

Eval design

Jobbeschreibung

About Luminary

Luminary is the AI-native system of action for wealth transfer services.

High net worth households controlling more than $90 trillion in assets are meaningfully underserved by technology that was built for the mass market. As private client investment and banking products have commoditized, these clients are looking for more tax and wealth transfer advice, advice that has previously been impossible to deliver at scale.

Luminary solves this problem by transforming complex estate planning and ownership documents into a structured, proprietary data asset that powers verticalized, agent-driven workflows for tax and wealth transfer services. Our built-in multi-firm collaboration connects a client's full advisor ecosystem, and is the core infrastructure that enables our customers to deliver a highly personalized, end-to-end wealth transfer service offering at scale.

Our 1,000+ customers are leaders across the private client services ecosystem: wealth managers from exclusive multi-family offices through top private banks and trust companies, national tax advisory practices, and law firms among the AM Law 100. Our focus and proprietary data position us to expand from a wedge in wealth transfer advice into large adjacent markets including trust administration and estate distribution.

Founded and led by CEO David Barnard, former Head of Wealth Management at AllianceBernstein, Luminary is backed by 8VC, Fin Capital, Rockefeller Capital, Focus Financial Partners, and several family offices.

How we work

Every engineer ships to production multiple times a day. We make small, reviewable changes that go out through CI as soon as they're ready.

Modern tooling. Meticulous , Amp , Honeycomb , Temporal , PostHog , Braintrust . We keep the team small and invest in tooling for leverage.

Engineers own outcomes. You'll talk to customers and feel their wins and pain points, sit with our in-house subject matter experts, and help us build a great business.

We support both remote and in-office preferences. We have an office in NYC near Bryant Park and a lively in-office culture, but if remote works better for you, you'll be in good company.

The Opportunity

The world of legal documents is perfect for AI: pages and pages of dense, high-value, unstructured data with lots of nuance. We often see our customers modeling households that have 60+ legal documents of all kinds, many of which interact with each other and have a number of amendments stacked on top of them.

We've built a very interesting pipeline to take those documents and turn them into a household-level knowledge graph: a mapping of the people, trusts, businesses, and holdings of a family represented in our domain-specific ontology. We use that knowledge graph as a contextual layer between the structured data on our platform (which we use for reporting and modeling) and the unstructured source of truth documents. It allows the various agents that we run on the platform to traverse across the whole plan and find the exact paragraph and context that they need.

This is the foundation, but it's not the end state: our customers are always asking us to do more and more - both to capture more nuance and improve our accuracy, but also to build more functionality on top of our application layer to take advantage of all the data and context we've gathered.

What you'll do

• Improve accuracy on our core extractions. Documents can be long, amended in unanticipated ways, or have unclear language. You'll work with our JDs and other experts to find the failure modes, figure out what's happening, fix them.
• Own our evals. The only way to maintain a system like this is through tests and evals. We have a strong foundation for evals, but think that it can be better and want to invest heavily here. You'll design and maintain the suites that tell us whether a change helped and to prevent regressions.
• Improve our shared agent harness. Tooling, parallel tool calls, context strategy, and the orchestration behind our agentic advisory workflows.
• Design for human review. Our users don't want a black box. Work with product and design on how a suggestion gets surfaced, explained, and accepted or rejected, then feed the corrections back into the system.
• Be informed on the state of AI. Models, capabilities, and usage pattern best practices shift weekly and it's our job to be aware of what's out there so we can remain at the frontier of the industry. You'll stay on top of the latest developments and, leveraging our evals, take advantage of the rising tide of models!

Who you are

Four or more years of software engineering experience, at companies with strong engineering practices. Beyond that:

An engineer who works on AI, not a researcher. You like shipping to production systems. You’re interested in and have experience with prompt and context engineering, eval design, and the unglamorous work of making probabilistic systems dependable.

Curious about the domain. There's so much nuance in