Applied AI Scientist/Engineer

Applied AI Scientist/Engineer — Training LLMs for Cybersecurity

Join Averlon's AI research lab, built to lead research at the frontier of cybersecurity. As attackers increasingly use automation and AI, the security community needs models that can reason about systems, find weaknesses, and help defenders respond faster.

Our lab trains and evaluates large language models (LLMs) across both defensive and offensive security. On the defensive side, it includes triage, mitigation, remediation, detection engineering, and digital forensics. On the offensive side, this includes penetration testing, red teaming, vulnerability discovery, and exploitation. We host our task experts on inference platforms, models specialized for individual security tasks, for use by other companies as well as internally.

As an Applied AI Scientist/Engineer, you will help shape Averlon's research agenda and model capabilities. You will train, evaluate, and benchmark LLMs on real-world security tasks, design reinforcement learning environments, and help publish findings that advance the field.

About the Team

The lab aims to publish one to two research papers and technical reports each month, including work on new reinforcement learning frameworks.

Research at Averlon spans LLM post-training involving continual learning, reinforcement learning, as well as agentic tool use, evaluation design, and large-scale benchmarking. You will work closely with our research team and alongside engineers with deep security expertise, who will help you learn the cybersecurity domain on the job.

Key Responsibilities

• Benchmark open-weight and proprietary LLMs on hundreds of offensive and defensive security tasks.
• RL of LLMs for security use cases as a post-training or continual learning process. This involves designing RL environments and reward signals for security tasks, including agentic settings that involve multi-step tool use.
• Track the research literature, reproduce relevant results, and assess new methods for adoption.
• Document results, negative as well as positive, to support internal decision-making.
• Contribute to research papers, technical reports, and blog posts that communicate the lab's findings to the wider community.

Required Skills and Qualifications

• Strong grounding in deep learning, evidenced by coursework and hands-on projects in which you trained DNNs or fine tuned LLMs/VLMs.
• Working knowledge of transformer architectures, attention mechanisms, and language (or vision) models training.
• Proficiency in Python and in PyTorch or a comparable deep learning framework.
• Strong analytical and problem-solving skills, with the ability to design rigorous experiments and draw sound conclusions from them.
• Clear written and verbal communication skills, with the ability to explain complex technical results to researchers and engineers.

Prior knowledge of cybersecurity or compute clouds is not necessary. You will learn the domain on the job, working alongside our team.

Preferred Qualifications

• Hands-on experience with LLM post-training or continual learning.
• Experience applying reinforcement learning, particularly to language models.
• Publications at venues such as NeurIPS, ICML, ICLR, ACL, or EMNLP, including workshop papers and preprints.
• Open-source contributions or software projects that others have adopted.
• Experience running and debugging distributed training on multi-node GPU systems.

Education and Experience

• An MS or PhD from the IITs, ISI, IISER, IISc, BITS Pilani, the IIITs, or top NITs is preferred.
• MS and PhD graduates do not need prior professional experience.
• Undergraduate candidates are encouraged to apply if they meet the required qualifications and have hands-on deep learning experience beyond coursework, such as research projects, internships, or publications.