Founding Ai Engineer

Nautilus Search; München; On-site; EUR 90,000 - 150,000; Full time

Job summary

Nautilus Search is hiring a Founding AI Engineer to own the layer that makes intelligent agents work in production. You will build and deploy models, manage evaluations, and ensure reliable operations across ERP/TMS/EDI integrations, in a founder-led Munich startup.

You will work directly with the founders to shape the engineering direction, recruit and guide the early team, and iterate quickly in a production environment with real customers.

Qualifications

• Experience running LLM systems in production with real users.
• Strong understanding of agents, tool use, structured outputs and context engineering.
• Hands-on with observability and tracing tools (Langfuse/LangSmith).

Responsibilities

• Own AI infrastructure behind the product: model selection, testing, fine-tuning and deployment.
• Set up evaluations and trace analysis across agents and sessions.
• Ensure production reliability: guardrails, retries, fallbacks, state/context management, escalations.
• Balance cost, latency and accuracy in production decisions.
• Shape the agent architecture and platform across backend services, data and queues.
• Contribute to technical direction and hiring for the growing team.

Skills

LLM systems production

Agents & tool use

Observability & tracing

Python/TypeScript

Asynchronous systems

Docker/CI/CD

Tools

Langfuse

LangSmith

Job description

Most AI agents never leave the demo. These ones run live inside real operations every day, reading inboxes, handling phone calls, and taking action across ERP, TMS and EDI systems for logistics companies across DACH. When they get something wrong, a real business feels it.

A funded, founder-led AI startup in Munich is hiring its Founding AI Engineer to own the layer that makes that work: the models, the evaluations, the observability, and the platform the rest of the engineering team builds on.

What you’ll own

• The AI infrastructure behind the product: choosing, testing, fine-tuning and deploying models, including open-source, and deciding when to switch


• Evaluations and trace analysis across agents and sessions, so every model change, prompt update or architecture shift is measured, not guessed


• Reliability in production: guardrails, retries, fallbacks, state and context management, and human escalation when the agent shouldn’t decide alone


• Cost, latency and accuracy, balanced deliberately rather than by accident


• Agent architecture and the core platform across backend services, integrations, data and event queues


• Technical direction alongside the founders, plus a real hand in hiring and shaping the engineering team as it grows

This is hands-on. You write production code daily. It’s not a research role, and it’s not an architecture role where you stop building.

What you bring

• 1.5 to 2+ years running LLM systems in production with real users, where you watched live traces and fixed what broke


• Strong grasp of agents, tool use, structured outputs and context engineering


• Hands‑on experience with observability and tracing tools such as Langfuse or LangSmith


• Solid engineering fundamentals: Python, TypeScript, SQL, APIs, data modelling and system design


• Comfort with asynchronous systems: queues, retries, idempotency, background jobs, failure handling


• Docker, CI/CD, cloud infrastructure and production debugging


• You follow model releases and research closely and turn them into decisions, not just opinions

Bonus: enterprise integrations, voice, email automation, or any logistics or supply chain exposure.

Who thrives here

You take ownership of messy, important problems without waiting for a spec. You ship the simple version first and know when it needs rebuilding properly. You build with AI tools as the default way you work. You’re direct, low ego, and happy to be close to customers and production incidents.

The setup

• Munich office, in person, five days. Not remote, and not a 9 to 5


• Small founding team working directly with the founders


• A fast process: one conversation, one hands‑on technical session building with an AI assistant, one founder conversation. Decision within a week