Senior AI Engieer - Voice & Agentic Systems

Xebia is a global AI-first, digital transformation, and engineering partner with over 25 years of experience and a team of 5,000 professionals across 16 countries. In the CEE region, they deliver modern applications, data platforms, and AI solutions for clients such as Millennium, ING, and FedEx. The company specializes in Artificial Intelligence, Data and Cloud, Intelligent Automation, and Digital Products.

• Proven experience shipping GenAI and agentic systems to production.
• Hands-on experience with LangGraph designing and running production workflows.
• Strong Python skills including asynchronous programming and building production services such as FastAPI.
• Strong software engineering and system architecture skills.
• Senior or lead-level engineering experience with ownership of significant technical decisions.
• Strong product sense with influence on what gets built.
• Understanding of production-quality software development, testing, observability, and reliability.
• Fluent English communication skills.
• Willingness to participate in regular in-person team sessions.
• Practical experience using AI-powered assistants like Claude Code, GitHub Copilot, or Cursor.
• Work from the European Union region with a valid work permit.

Strongly preferred:

• Experience building voice agents or real-time voice pipelines.
• Experience with Pipecat, LiveKit Agents, or similar frameworks.
• Practical experience with streaming STT/TTS technologies such as Deepgram, Soniox, ElevenLabs, or Cartesia.
• Experience with voice activity detection, turn-taking, interruption handling, and voice latency optimization.
• Production experience with WebRTC and/or LiveKit.
• Experience with telephony technologies like Twilio, Telnyx, or SIP.

Nice to have:

• Experience with LLM evaluation and observability tooling such as Langfuse, LangSmith, or custom frameworks.
• Experience working with model providers such as OpenAI, Anthropic, and Gemini.
• Experience routing between models based on cost, latency, and capability.
• Experience with MCP or similar tool-integration patterns.
• Knowledge of TypeScript and React to contribute to client-side voice interface.
• Experience in startup or 0-to-1 product environments.
• Experience applying GenAI within SDLC including workflows and prompt patterns.
• Interest and familiarity with AI-driven practices such as agent-based workflows and AI-augmented development.

Agent engineering:

• Design and build agent workflows using LangGraph including state management, tool and function calling, multi-step reasoning, memory, error recovery, and human handover.
• Define agent decision-making during live conversations balancing latency and interruptions.
• Build and maintain backend integrations enabling agent actions.
• Ensure reliable agent workflows for real-world production.

Voice pipeline:

• Build and tune real-time voice pipelines using technologies such as Pipecat, LiveKit Agents or similar.
• Work with streaming STT, TTS, voice activity detection, turn-taking, and interruption handling.
• Improve conversation quality via end-of-turn detection, barge-in handling, filler and back-channeling, and error recovery.
• Measure and reduce end-to-end latency.
• Ensure reliable voice interactions under real-world network conditions.

Architecture and quality:

• Shape system architecture and decide on build-vs-buy for models and speech technologies.
• Build evaluation workflows using conversation-level test sets, regression suites, and LLM-as-judge methods.
• Implement observability and tracing with tools like Langfuse and LangSmith.
• Diagnose production issues via metrics and real user sessions.
• Apply engineering best practices: code review, automated testing, CI/CD, documentation, maintainable design.

Product:

• Translate user needs into technical proposals.
• Challenge assumptions and suggest alternatives to improve user experience.
• Prototype rapidly and harden solutions for production.
• Communicate technical trade-offs clearly to technical and non-technical stakeholders.
• Participate in regular in-person planning, design, and review sessions.

Work from the European Union region and a valid work permit are required. Participation in regular in-person team sessions is expected. Open to collocating in Berlin, with Madrid or Barcelona as alternatives.