Engineering
Senior ML Engineer
Sofia, BulgariaFull-time
Zeta Cortex is a company building software that secures enterprise agentic AI, giving organizations control over what their AI agents are allowed to do.
As a Senior ML Engineer you'll build the models at the core of that work, from training data to production, in a small team where you own problems end to end.
What you'll do
- Build and maintain training and evaluation datasets from real-world security data.
- Fine-tune and evaluate open-weight language models for security classification tasks.
- Design offline evaluations and regression checks so every model change is measurable.
- Take models to production under tight latency and reliability requirements.
- Build feedback loops that turn human review into better training data.
- Work closely with the engineers building our platform, from backend integration to how model results reach users.
What you bring
- Hands-on experience fine-tuning and evaluating open-weight language models, and taking at least one to production.
- Strong Python and the usual stack (PyTorch, Hugging Face), and comfort working in a production codebase.
- Experience turning messy real-world data into reliable datasets.
- A security mindset: you've thought about how an attacker would get around a classifier.
- You like owning a problem end to end in a small team.
Nice to have
- Security ML background, such as malware, intrusion or fraud detection, or adversarial ML.
- Low-latency inference experience: quantization, vLLM, ONNX Runtime or llama.cpp.
- Familiarity with AI agent frameworks, MCP or tool-calling models.
- Go or Rust.
Why Zeta Cortex
- An urgent problem: enterprises are giving AI agents real access to their systems.
- Hard, real-world security problems rather than benchmarks.
- Real ownership in a small team.
- Work closely with top experts in the field and learn a lot along the way.
- Work at the edge of technology and innovation in a field that's still being defined.
How to apply
Email careers@zetacortex.ai with your CV or GitHub and a few lines about a model you've taken to production: what it did, how you evaluated it and what you'd do differently.