About the role
You'll work on the intelligence layer of a security platform, the part that reads raw scanner output and turns it into findings a human can act on. That means working with AI models, prompts, and figuring out whether the answers are actually any good. This is a beginner-friendly role. If you've built something with an AI API and gotten curious about why it worked sometimes and not others, that's the right starting point.
What you'll do
- Work on AI features that read security scan results and explain them in plain language
- Write and improve prompts, then test whether your change actually made things better
- Build test sets: collect examples, check the model's answers against them, and track the score
- Help cut down false alarms. Teach the system to spot a real problem versus noise
- Help connect models to the tools and data they need through MCP servers
- Clean and prepare data: parsing tool output, fixing messy JSON, organising examples
- Learn how retrieval works (finding the right information to give the model) and help improve it
- Keep notes on what you tried and what happened, including the things that didn't work
What we're looking for
- Python: you can write a working script on your own; you don't need to be an expert
- Basic understanding of what machine learning is: that a model learns from data, and that testing it on data it already saw doesn't prove anything
- You've used an AI chatbot or API enough to notice it gets things wrong, and you're curious about why
- Comfortable reading documentation and looking things up; this field moves faster than any course
- Patient and honest. A lot of this job is trying an idea, measuring it, finding out it didn't help, and saying so
- AI tools are welcome, but you must learn, explain, test, and take responsibility for any code you submit. Do not use generated code you do not understand
You don't need to already know these. We'll teach you
- How LLM APIs, prompts, tokens, and context windows work in practice
- Building evaluation sets and scoring model output
- Retrieval and vector search (RAG)
- AI agents and multi-step workflows
- Running AI models locally
- The security knowledge behind the findings
Nice to have
- MCP (Model Context Protocol): a general idea of what it is and how it lets AI models connect to external tools and data. Having built or run one is a strong plus.
- Fine-tuning: a basic understanding of what it means to train a model further on your own data, and roughly when it's worth doing versus just improving the prompt. Hands-on experience (LoRA, Hugging Face) is a bonus.
- Any coursework or self-study in ML, including Andrew Ng's course, fast.ai, or a university unit
- Familiarity with pandas, NumPy, scikit-learn, or Jupyter notebooks
- SQL: enough to pull data out of a database
- JavaScript or Go: useful for the parts of the stack around the model
- A project we can look at: a notebook, a small app, a Kaggle entry, a blog post
- Any interest in security, even with no background in it
What you'll get
- Real production AI work, not a toy chatbot. Your changes affect what users see
- Mentorship from day one; you'll be taught the tools, not handed them