AI Engineer (m/f/d) - Focus: Agentic Systems & Backend Engineering (application programmer)

Germany | Oct. 15, 2026

Report as Closed

Company: conmeet GmbH

Country: Germany

Type: Hybrid remote

Employment: Full-time

Description: Your role

You develop our AI systems - from agent-based assistants to a platform on which our customers can put together their own agents, tasks and processes.

The role is hands-on: You write code, design agent architectures and operate what you build yourself. You work at the point where it is decided whether an AI feature is only convincing in the demo or in the everyday life of our customers works reliably.

You work closely with backend development, product and [other interfaces] and actively contribute your assessment to architecture and product decisions.

Our team and your opportunity

smaller 25 employeesLocations: 2Software development

A young, dynamic team awaits you at conmeet that works with passion and expertise on the future of digital craftsmanship. We maintain a relaxed atmosphere in which fun at work is not neglected, without losing sight of our ambitious goals. Teamwork is very important to us: We support each other, share knowledge and celebrate our successes together.

We believe in the continuous development of our team members. That’s why we encourage and challenge you in equal measure. You get the opportunity to grow with complex, challenging tasks and take your skills to the next level. Regardless of whether it is about solving technical problems, developing onboarding processes or building customer knowledge - with us you will always find new challenges that will advance you as a customer success professional.

Your main tasks

- (further) development of our agent-based AI assistant - from conception to implementation to operation
- Co-creation of a dynamic multi-agent system in which customers create their own agents, tasks and workflows themselves and configure
- Design of agent systems whose behavior and quality are controllable and comprehensible - including guardrails, authorization concepts and human-in-the-loop mechanisms
- Development of systematic evaluation strategies: You make it possible to check whether changes to prompts, tools and agents are actually improvements - instead of just "feeling better"
- Integration of AI and ML concepts into an existing, grown SaaS application - with everything This includes: data model, authorizations, performance, costs
- Operation and optimization: observability, tracing, latency, cost and stability optimization

Your profile

Understanding of LLM and ML

- Very good conceptual understanding of how LLMs work - and the limits -
- Understanding of other ML concepts and how they can be meaningfully integrated into business logic

In-depth understanding from:

- Tokenization/Tokens, Embeddings and Context Windows
- Vector Search, Retrieval and Re-ranking (RAG)
- Hallucinations, Context Bloat and other typical error modes - including strategies to detect and mitigate these
- Tool Calling and Structured Outputs
- A2A (Agent2Agent Protocol), MCP (Model Context Protocol) and Human-in-the-Loop
- Security risks of LLM and agent systems (including prompt injection, tool misuse, data leakage)
- Prompt engineering – formulation, structure and control of individual model requests
- Context engineering – targeted selection, preparation and management of what ends up in the context window
- Harness engineering – design of the environment in which an agent operates: tools, feedback loops, control flow and termination conditions
- ...as well as a clear one Understanding how these three levels differ and when which is the right lever

Agent architecture

- Understanding of multi-agent architectures, agent handoffs and orchestration (A2A)
- Experience with guardrails and permissions, especially RBAC and ABAC in user-facing SaaS systems0
- Experience or solid understanding of asynchronous processing, background jobs and task queues

MLOps / AgentOps

- Ability to develop systematic evaluation strategies for agents and other ML components
- Understanding of quality metrics and evaluation methods – e.g. B. LLM-as-a-Judge, Task Success Rate, Tool Call Accuracy
- Testing and versioning of agents, prompts and tool definitions
- Optimization of quality, stability, latency and costs
- Observability, logging and tracing

Mindset

- Genuine interest in testing different AI and ML concepts in an existing SaaS application and making them productive
- Pragmatism: You know the difference between an impressive demo and a system that also works well with the 10,000. still does the right thing

Techstack

This is how we work:

What is printed in bold on the left is our stack - what is next to it shows that we also take experience with comparable technologies into account. Experience with as many of these points as possible is an advantage, but we do not expect the complete list:

LLM / AI: Anthropic API – Anthropic Client, Anthropic on Vertex AI (or comparable LLM clients and APIs)

ML toolchain: Development and evaluation of your own ML applications, e.g. with PyTorch, TensorFlow, scikit-learn, NumPy, pandas, CUDA

Cloud: Google Cloud Platform – Vertex AI, Cloud Run, GKE (or comparable platforms such as AWS or Azure)

Infrastructure: Kubernetes, Docker

Backend: Python, Django

Database: PostgreSQL (or a comparable RDBMS). Hybrid, or on site to...

Home office:
Scope: By arrangement

Apply here:

Web: Apply here

Emails: