Erstellt am 15. Mai 2026
AI Engineer - RAG and Agentic AI (m/f/d)
Advantest
Böblingen, Baden-Württemberg 71034, Germany
Vollzeit
Reference: 2119980420
- Design, implement, test, and continuously optimize end-to-end RAG pipelines, including data parsing, ingestion, prompt engineering, and chunking strategies.
- Curate and develop high-quality datasets, using synthetic data generation for robust training and evaluation.
- Rigorously evaluate LLM applications on metrics including correctness, latency, and hallucination.
- Assist in the deployment of LLM-based applications, analyze user feedback, and contribute to iterative improvements.
- Write clean, maintainable, and testable code following best practices.
- Collaborate with cross-functional teams to integrate AI components into other systems.
- Master's or Ph.D. in Computer Science, Machine Learning, or a related field and a minimum of 2 years of hands-on industry experience in software engineering.
- Experience operating RAG systems in production environments, including monitoring, debugging, and continuous improvement based on real user behavior.
- Solid understanding of software engineering practices applied to AI systems (testing, CI/CD integration, versioning, and reproducibility).
- Ability to balance research innovation with long-term maintainability and customer-ready quality standards.
- Clear communication and presentation skills.
Good To Have:
- Experience with observability stacks (e.g., Prometheus, Grafana, OpenTelemetry) applied to AI or backend services.
- Familiarity with enterprise deployment constraints such as air-gapped systems, license compliance, and distribution of AI-enabled software to customers.
- Exposure to agent frameworks, tool-calling patterns, or multi-step reasoning architectures.
- Hands-on experience with vector databases (e.g., Milvus) and modern RAG architectures, such as Graph-based Retrieval-Augmented Generation.
- This role emphasizes long-term ownership of Retrieval-Augmented Generation systems as a core product capability, not just experimentation with large language models.
