About the role
You will design and ship AI features for clients and for SOWTK: assistants, document processing, search and workflow automation built on large language models. You will make them accurate, safe, measurable and affordable in production.
What you will do
- Build applications on large language models through their APIs
- Design retrieval (RAG): chunking, embeddings, re-ranking and vector search
- Build tool-using agents and workflow automation
- Set up evaluation suites for quality, groundedness and regressions
- Add safety controls: prompt-injection checks, output validation, data privacy
- Deploy, monitor and control the cost of AI services
Minimum qualifications
- Bachelor's degree in Computer Science, AI or a related field, or equivalent practical experience
- 5+ years of software engineering, including 2+ years shipping LLM features to production
- Production Python or TypeScript
- A RAG or agent system you built and can demo or show in code
- Evaluation suites you built (RAGAS, Promptfoo or custom)
- Work to the OWASP Top 10 for LLM Applications
- English at CEFR C1 or above (IELTS 7.0 or equivalent)
Expert knowledge
- Retrieval quality: chunking strategy, hybrid search, re-ranking and citation grounding
- LLM evaluation: test sets, LLM-as-judge calibration and regression tracking
- Agent design: tool schemas, planning, error recovery and human approval steps
- LLM security: prompt-injection defence, output validation and data isolation (OWASP LLM Top 10)
Preferred qualifications
- Model Context Protocol (MCP) servers or tools
- Published projects, papers or open-source AI work
- Arabic NLP
- AWS Machine Learning Specialty or Azure AI Engineer (AI-102)
Before any offer we verify your certificates, qualifications, references and right to work in the UK. Security roles also need background screening (BPSS) and may need security clearance.
SOWTK is an equal opportunity employer. We welcome applicants of every background and do not discriminate on any characteristic protected by the Equality Act 2010.