Senior AI Engineer

NXT; Berlin; Hybrid; EUR 110.000 - 140.000; Vollzeit

Benefits dieser Stelle

Hybrid work model

Zusammenfassung

NXT is building AI operations for regulated mid-market clients in the DACH region, delivering production-grade AI systems with governance. We are expanding the AI platform to include document understanding, extraction, and agentic components that run in real client engagements.

The Senior AI Engineer will own LLM pipelines, evaluation, and tool-based architectures, with a focus on observable, governable systems that ship into production.

Qualifikationen

• Strong backend engineering background • Python at depth and production systems ownership
• LLM systems in production: designed, deployed, operated, and evaluated
• Tool-based agent architectures, structured extraction, retrieval, and eval
• Cloud-native delivery: containers, CI/CD, Postgres, observability
• Experience in regulated environments (banking/healthcare) and production-grade norms
• Business-fluent German is a strong plus; client environments are German-speaking

Aufgaben

• Design and build the AI layer of the platform: document understanding, extraction, classification, and structured decision logic
• Develop agentic components with guarantees and traceable decisions
• Create reusable platform components from engagements and collaborate with Forward Deployed Engineers
• Define a rapid, production-first development workflow with agents as standard
• Ensure cost, latency, and quality trade-offs are optimized in real-case volumes

Kenntnisse

Python backend

LLM production

Tool-based agents

Cloud-native delivery

Regulated environments

Tools

Postgres

Celery

Go

TypeScript

React

Claude/Vertex AI

Jobbeschreibung

I left McKinsey in March to build again.

After years of business building and GenAI work in regulated mid‑market environments, one thing became clear: the bottleneck in AI is not capability. It is turning capability into reliable systems that operate inside real businesses. Most companies experiment with AI. Very few build systems that are governed, production‑grade, and embedded in operational workflows.

NXT closes that gap. We are an AI Operations company — we build and run AI systems for regulated mid‑market companies and PE portfolios in DACH. Not consulting: production systems with operational accountability, deployed into finance, back office, and operational workflows, creating measurable value in weeks.

Five months in, this is no longer a thesis. We have anchor clients in banking and healthcare, seven‑figure revenue in year one, a team of seven, and ISO 27001 certification underway — fully bootstrapped, funded by client revenue from day one.

To build the AI layer behind this — document understanding, decision logic, and the agentic components that run real cases — I am looking for a Senior AI Engineer.

How We Build

We do not build the platform in isolation. We build it inside real, paying client engagements in regulated industries — currently banking and healthcare — and we own the reusable IP that emerges. The first engagements are the birthplaces of the first platform components. This is a deliberate choice. The platform runs in production from day one, in regulated environments, with real operational data and real consequences. For the Senior AI Engineer this means: your pipelines are never evaluated on a benchmark. They are evaluated on last week's cases at a bank, with a control function reading the logs.

The Role

This is a senior individual contributor role on the platform team. You build the AI layer of NXT Core: the pipelines that read messy documents, the logic that turns them into structured decisions, and the agentic components that act on them under strict boundaries. You are not insulated from clients — when your system meets real data for the first time, you are in the room. You work directly with the founder, the Director of Engineering, and the Forward Deployed Engineers who carry your components into production. This is a builder role. You merge code every week, you own the quality of what ships, and you set the technical bar for how AI is engineered at NXT. For the right profile this is a Staff role: the person who shapes the system, not just contributes to it.

What You Will Do

LLM pipelines in production: Design and build the AI layer of the platform — document understanding, extraction, classification, and structured decision logic across real‑world inputs. Own model choice, prompt and tool design, and the cost, latency, and quality trade‑offs on real case volumes. Build evaluation infrastructure that catches regressions before a client does.

Agentic components with guarantees: Build tool‑based agent architectures where the model proposes and deterministic logic executes. Define what the model may decide, what needs a rule, and what needs a human. Make every decision traceable, so the system can be explained to a regulator or an internal control function.

Platform components that compound: Turn what worked in one engagement into reusable components in NXT Core — extraction pipelines, evaluation harnesses, approval flows. Work closely with the Forward Deployed Engineers so the platform absorbs what they learn in the field instead of solving the same problem twice.

AI‑native way of working: Coding agents are part of the default development model. Help define how a small team ships fast without losing the guarantees the system has to hold in production.

Technical Challenges You Will Own

• Document understanding at production quality across messy, real‑world inputs — scanned forms, emails, attachments, legacy exports
• Evaluation and regression infrastructure for LLM‑based systems that an auditor can read
• Safe execution boundaries between deterministic and agentic logic
• Tool‑based agent architectures with retries, partial execution, and approval logic
• Making LLM‑based systems observable, governable, and debuggable in production
• Cost, latency, and quality trade‑offs at real case volumes
• An engineering environment where coding agents are part of the default development model

What You Need

• Strong backend engineering background — Python at depth; you have designed, built, and shipped production systems end‑to‑end, and you still carry core components yourself
• LLM systems in production: designed, deployed, operated, and evaluated — beyond experimentation and demos
• Hands‑on with tool‑based agent architectures, structured extraction, retrieval, and systematic evaluation
• Cloud‑native delivery — containers, CI/CD, Postgres, observability
• Experience in or genuine respect for regulated environments (banking, insurance, healthcare) and what production‑grade means there
• Comfort stepping into client engagements when your system meets real data — this is not an insulated research role
• Business‑fluent German is a strong plus — our client environments are German‑speaking

Who This Is For

It is not a fit for research‑focused ML backgrounds without shipping track record, data science or analytics profiles without backend depth, low‑code AI builders, or engineers who want a stable product with clean requirements. We are early. This is Aufbauarbeit — deliberate, funded, and already in production, but Aufbauarbeit.

Stack

Python, FastAPI, Postgres, Celery, and Go on the backend. TypeScript and React on the front. Claude via Vertex AI in EU (Frankfurt) for the LLM layer. Coding agents are part of the default development workflow.

Practicalities

Düsseldorf is our center of gravity; Berlin works with regular presence at clients in NRW. Hybrid by default. Competitive base plus bonus. Start as soon as your notice period allows.