AI Developer with Python for Customer Care AI Platform team (hybrid) gesucht in Berlin

As an AI Engineer on this team, you will build the core intelligence systems behind our multimodal AI platform.You will be responsible for moving beyond simple chat interfaces to build high-performance, real-time systems that handle complex reasoning, deep context retrieval, LLM orchestration, retrieval-augmented generation (RAG) and seamless voice interactions., * Design Agentic Workflows: Design and implement LLM-based systems that go behind response generation - enabling structured tool usage, workflow orchestration, and secure interaction with internal services via MCP (Model Context Protocol).

• Build and Optimize RAG & CAG: Develop high-performance Retrieval-Augmented Generation and Context-Augmented Generation pipelines to ensure accurate, relevant, and low-latency responses. Continuously improve context management, ranking strategies, and grounding mechanisms to support complex, multi-step interactions.
• Voice Channel Mastery: Develop and optimize real-time Speech-to-Speech (S2S) pipelines, focusing on streaming architectures, latency reduction (including Time to First Word - TTFW) and maintaining a natural conversational flow.
• Evaluation, Quality & Alignment: Build and maintain an automated QA module, including LLM-as-a-judge patterns, to measure accuracy, safety, latency, and resolution quality at scale. Translate evaluation insights into systematic models and prompt improvements..
• Model Strategy & Hybrid Integration: Integrate and operate both commercial foundation models (e.g., OpenAI, Anthropic, Google) and open-source alternatives (e.g., Qwen, Kimi, DeepSeek, Moonshot, GLM), selecting and optimizing models based on performance, latency, cost, and use-case requirements.

• Strong Python and/or Java Engineering Skills: Advanced-level Python development experience, including asynchronous programming (e.g., FastAPI, asyncio) and building high-performance, production-grade services. Experience with streaming architectures is a strong advantage.
• LLM Application & Multi-Agent Orchestration Experience: Hands-on experience building LLM-powered systems, including multi-step workflows, stateful agents, and tool invocation. Familiarity with orchestration frameworks such as LangChain, LlamaIndex, or LangGraph, particularly in building stateful, multi-turn agents.
• Advanced Retrieval & Context Management: Deep understanding of vector databases (e.g., Weaviate, Qdrant, pgvector, Elasticsearch), semantic search, embedding strategies, and re-ranking techniques. Experience designing and optimizing RAG pipelines.
• Real-Time & Low-Latency Systems: Experience in designing systems that operate under latency constraints, including streaming APIs, event-driven architectures, and performance optimization. Understanding of trade-offs between quality, cost, and response time.
• Evaluation-Driven Development: Experience in implementing evaluation frameworks for LLM-based systems, including automated QA pipelines and LLM-as-a-judge patterns.
• Familiar with API Design: knowledge of RESTful API design, OAuth2

Our mission is to build a modern ecosystem used for all IONOS customer support needs. The tools developed by us are used in over 20 locations, by more than 2.000 users, supporting 8 million customer contracts in 10 markets. The development team has full responsibility for the development lifecycle. This means we plan, develop, test and deploy our software without any other internal or external dependencies. Our portfolio revolves around an internally built CRM which is now being enhanced with AI capabilities. About the product you will be building: We are building a next-generation AI platform designed to redefine how our company interacts with customers. This isn't just a chatbot; it's a high-performance, multimodal AI ecosystem powered by state-of-the-art Speech-to-Speech (S2S) models, advanced Large Language Models (LLMs), and intelligent orchestration frameworks. Our platform will understand, reason, and respond across text and voice - while seamlessly executing real-time actions to resolve customer needs. We are aiming for a hybrid architecture of Open Source LLMs, industry-leading proprietary models, and Model Context Protocol (MCP) to enable contextual reasoning, tool invocation, and seamless orchestration across systems. The goal is not just to talk to the customer, but to act on their needs. What makes this project unique: The Voice Frontier: We are building low-latency, emotive speech-to-speech pipelines for a truly natural voice channel experience. Deep System Integration: Our platform connects directly to the company's core systems via MCPs, allowing the AI to access real-time customer context and execute complex workflows. Self-Evolving Logic: We are developing an automated QA and evaluation module that continuously analyzes interactions across channels.By programmatically measuring quality, accuracy, latency, and resolution outcomes, we can close the feedback loop, and adapt system behavior in hours, not weeks. Hybrid Innovation: You'll work at the intersection of "build vs. buy," integrating the best of the open-source community with custom-built internal infrastructure. What's in it for you: You won't just be shipping code; you'll be part of making this concept evolve and shift. You'll join a friendly, experienced team where your voice matters and your contribution shapes real-world outcomes. You'll work in a modern environment with technologies and practices that help us ship reliable software efficiently.