Senior Researcher, Neurosymbolic reasoning

Senior Researcher – Neurosymbolic Reasoning

About the Role

We are seeking a Senior Researcher specialising in neurosymbolic reasoning, explainable AI, and AI safety to join an advanced research team in Helsinki.

The team develops next-generation AI systems for content understanding, multimodal reasoning, model safety, alignment, interpretability, and responsible deployment. The role focuses on advancing state-of-the-art research while ensuring that promising methods can be translated into scalable, production-ready systems.

You will work at the intersection of large language models, vision-language models, agentic AI, knowledge representation, neural-symbolic methods, and AI safety.

Key Responsibilities

• Design and implement novel architectures for AI systems capable of understanding and reasoning over complex multimodal information, including text, images, video, and contextual or cultural knowledge.

• Research advanced methods for visual question answering, visual reasoning, text-based reasoning, and multimodal knowledge grounding.

• Explore knowledge graphs, structured representations, and neurosymbolic techniques to improve model reasoning and factual grounding.

• Develop methods to improve the interpretability, transparency, robustness, and controllability of large-scale AI systems.

• Research alignment techniques for large language models and agentic AI systems, including reinforcement learning from human or AI feedback and other preference-based alignment methods.

• Investigate alignment and reliability across chain-of-thought reasoning, multi-step workflows, planning, and tool-using AI agents.

• Study the internal behaviour of neural networks using interpretability, mechanistic analysis, representation analysis, and related techniques.

• Develop and integrate explainable AI methods that provide meaningful insight into model decisions and internal representations.

• Develop data attribution techniques to estimate the influence of individual training examples or datasets on model behaviour and predictions.

• Research model editing, unlearning, continual learning, and controlled adaptation techniques.

• Build rigorous benchmarks and datasets for evaluating reasoning quality, reliability, transparency, interpretability, and model behaviour.

• Develop defences against prompt injection, jailbreaking, adversarial inputs, and attacks targeting AI planning or tool-use capabilities.

• Reproduce and evaluate recent research in AI safety, reasoning, interpretability, and multimodal AI.

• Translate promising research papers and experimental approaches into efficient, scalable implementations.

• Communicate research findings and technical recommendations clearly to research, engineering, and product stakeholders.

Essential Requirements

• PhD in Computer Science, Artificial Intelligence, Machine Learning, Mathematics, Deep Learning, or a related technical discipline.

• Strong research background in artificial intelligence with demonstrated work in areas such as explainable AI, AI safety, reasoning, or trustworthy machine learning.

• Strong knowledge of large language models, vision-language models, transformer-based architectures, and neural networks.

• Experience with neurosymbolic AI, graph-based reasoning, knowledge representation, or related structured reasoning techniques.

• Strong programming skills in Python and experience with Java, C++, or comparable languages.

• Hands-on experience with machine learning frameworks such as PyTorch or TensorFlow.

• Strong ability to design experiments, analyse model behaviour, and evaluate complex AI systems.

• Experience translating research concepts into practical implementations.

• Strong analytical and problem-solving ability.

• Ability to work effectively within international and interdisciplinary research teams.

Preferred Qualifications

• Research experience in AI alignment, model safety, robustness, interpretability, or controlled learning.

• Experience with reinforcement learning from human feedback, reinforcement learning from AI feedback, preference optimisation, or related alignment methods.

• Experience with mechanistic interpretability, representation analysis, activation analysis, probing, or neural network reverse engineering.

• Experience with model editing, machine unlearning, continual learning, or knowledge modification.

• Knowledge of multimodal reasoning and vision-language architectures.

• Experience with knowledge graphs, graph neural networks, symbolic reasoning, or hybrid neural-symbolic systems.

• Research experience in agentic AI, planning, tool use, or multi-step LLM reasoning.

• Experience evaluating or mitigating prompt injection, jailbreaking, adversarial attacks, and other model-security risks.

• Experience building datasets or benchmarks for AI safety, interpretability, reasoning, or reliability.

• Experience developing high-performance implementations of research methods for production or large-scale experimentation.

• Contributions that have introduced significant new methods, architectures, or capabilities within machine learning or artificial intelligence.

Research Track Record

A strong publication record at leading AI, machine learning, computer vision, or natural language processing conferences is highly valued.

Relevant venues may include:

• NeurIPS

• ICLR

• ICML

• AAAI

• ACL

• EMNLP

• NAACL

• CVPR

• ICCV

Candidates with recognised achievements in international computer science, mathematics, machine learning, or artificial intelligence competitions may also be of particular interest.

Candidate Profile

The successful candidate will combine strong academic research capabilities with the ability to develop working AI systems.

You should be comfortable investigating open-ended research questions, challenging existing approaches, building experimental systems, and translating theoretical advances into robust implementations.

The role is particularly suited to researchers interested in the intersection of advanced AI reasoning, multimodal intelligence, interpretability, model alignment, and AI safety.