LLM-Based Knowledge Extraction and Failure Analysis Internship

About the role

Siemens Research & Predevelopment (RPD) is the central R&D department of Siemens, focusing on future technologies for industry, infrastructure, mobility, and healthcare. We are seeking an Intern to support the Software Systems and Processes team in Princeton, NJ, by researching and developing scalable intelligent systems using Large Language Models (LLMs) and semantic technologies.

Responsibilities

• Design, test, and refine prompts and context-selection strategies that help models classify failures, use relevant evidence, and produce consistent structured JSON outputs.
• Analyze LLM output quality to understand why models choose incorrect failure classes, overlook important evidence, rely on misleading context, or generate inconsistent explanations.
• Create evaluation examples, test cases, scoring rubrics, and error-analysis summaries to measure classification accuracy, evidence quality, explanation quality, and robustness.
• Improve JSON schemas, validation checks, metadata fields, and intermediate representations used by downstream analysis and reporting workflows.
• Prototype improvements to data preparation, retrieval or context assembly, prompt templates, output formatting, post-processing, and evaluation logic in Python-based AI pipelines.
• Collaborate with software engineers, AI researchers, and domain experts to understand failure categories, edge cases, expected model behavior, and quality requirements.

Qualifications

• Currently enrolled in a Master’s or PhD program in Computer Science, Artificial Intelligence, Data Science, Knowledge Engineering, Information Science, or a closely related technical field.
• 3+ years of foundational knowledge and research or project experience in Artificial Intelligence, Machine Learning, Generative AI, NLP, Data Engineering, or knowledge-based intelligent systems.
• 3+ years of hands-on programming experience in Python, including experience with AI/ML libraries or frameworks such as PyTorch, TensorFlow, Hugging Face Transformers, scikit-learn, LangChain, LlamaIndex, or similar tools.
• Hands-on experience with prompt engineering, context engineering, structured LLM outputs, or LLM-based information extraction and classification workflows.
• Strong understanding of data modeling, structured outputs, metadata design, schema quality, validation concepts, and data quality principles.
• Experience designing, implementing, or evaluating AI workflows that combine LLMs with structured context, retrieval, information extraction, classification, or rule-based validation.
• Demonstrated ability to conduct independent research, critically analyze complex problems, work through ambiguity, and deliver structured technical outputs on defined timelines.
• Strong written and verbal communication skills in English, with the ability to explain technical concepts clearly to both technical and domain-expert audiences.

Preferred Skills

• Knowledge of transformer-based models, attention mechanisms, NLP/NLU methods, named entity recognition, relation extraction, question answering, or text classification.
• Experience building reproducible data or AI pipelines, including data ingestion, validation, testing, documentation, and workflow orchestration with tools such as Apache Airflow, Prefect, Git, Docker, or similar technologies.
• Ability to work with domain experts to translate engineering failure categories, business requirements, and quality expectations into clear prompts, evaluation criteria, and structured output formats.
• Excellent analytical skills, attention to detail, and ability to reason about model behavior, evidence quality, data ambiguity, reproducibility, and maintainability of AI pipeline outputs.
• Capacity to work independently, prioritize effectively, communicate progress clearly, and collaborate in an interdisciplinary research environment.
• Interest in applying LLMs, knowledge extraction, and quality-focused AI engineering to industrial software systems, intelligent automation, or enterprise-scale engineering use cases.