AI Researcher for AI Stack Security (m/f/n)
ESET
- Bratislava, Slovakia
- Mid Level
- Posted
Skills
Required
- large language models
- transformer architectures
- dataset preparation
- experiment design
- model distillation
- inference performance optimization
- quantization
- pruning
- latency- and memory-constrained environments
- Python
- AI/ML ecosystem
Preferred
- cybersecurity / threat research / adversarial machine learning
- endpoint / embedded hardware
- deployment-ready
- architecture trade-offs
- language model security and
You will become part of the newly established AI Threat Research and AI Conversation Security team within the Technology Division in Bratislava. As one of the founding members, you have a unique opportunity to shape the team's direction and methodology from the ground up.
Responsibilities
- The team researches attacks against conversational AI systems, probing and breaking them to understand how adversaries exploit large language models.
- Beyond research, the team designs and implements detection mechanisms and translates these defenses into shipping products across endpoint, cloud, and standalone form factors.
- As an Experienced AI Researcher, you research, train, and optimize the language models that detect attacks on conversational AI.
- Beyond core model work, you carry models from early experiments to deployment-ready detectors that run within strict performance budgets across endpoint and cloud environments.
- Research and prototype specialized language model detectors for malicious prompts and attempts to manipulate conversational AI, turning adversarial research findings into model designs.
- Train and fine-tune small language models for detection tasks, from dataset preparation through experimentation to production handoff.
- Distill large models into compact, specialized detectors that balance detection quality against strict latency and memory budgets.
- Optimize inference performance through quantization, pruning, and architecture trade-offs, keeping detection real-time even on endpoint hardware.
- Evaluate detector performance across diverse conversation datasets, measuring detection quality, robustness, and false-positive rates.
- Track state-of-the-art research in language model security and efficiency, deciding which advances are worth adopting.
- Collaborate with security researchers and product engineers to deploy detectors across endpoint, cloud, and standalone form factors.
Requirements
- Requirements Deep understanding of how large language models work internally — transformer architectures, attention mechanisms, and the trade-offs behind their performance and cost.
- Experience training and fine-tuning language models, from dataset preparation to experiment design.
- Hands-on experience with model distillation, turning large models into compact detectors that preserve quality.
- Experience with inference performance optimization, including quantization and pruning for latency- and memory-constrained environments.
- Rigorous approach to model evaluation, measuring quality, robustness, and false-positive behavior.
- Proficiency in Python and its AI/ML ecosystem.
- Strong drive for continuous learning and staying current with state-of-the-art research.
Preferred
- Requirements University degree in Machine Learning, Computer Science, Mathematics, or a related technical field.
- Background in cybersecurity, threat research, or adversarial machine learning.
- Experience running models on endpoint or embedded hardware.
Benefits
- Multisport card Travel card "električenka" Supplementary pension saving
- Sick days Family Marriage allowance
- Childbirth allowance Parental contribution ESET family events (Family day, St.Nicholas) Office Refreshments in the office (fruits, coffee, soft drinks) Let´s get together breakfast
- Christmas presents Discounts from external providers
Technologies
Python, PyTorch, Hugging Face, scikit-learn, PEFT, bitsandbytes, ONNX, Docker, Git, JIRA, Confluence Compensation Basic wage component (brutto): from 3500 EUR/m. (brutto) * The final basic wage component can be increased accordingly to individual skills and experience of the selected candidate. * Performance bonus 2 times per year up to 10% of the basic salary paid for the evaluation period(usually 6 months). #LI-JK1 #senior #medior #LI-Hybrid Benefits Health & well-being BenefitPlus Healthcare (check-ups/annual programs)
Activity room, quiet room ESET events (Žranica, Christmas party..), teambuildings Other Loyalty allowance Loyalty extra days off Loyalty cake Loyalty glass award Refer a friend
Primary location Bratislava Additional locations Time type Full time
