AI Engineer (Biometrics and Data Science)

• Design, build, and iterate AI agents for the department, for example ADaM code generation agent, SAP generation agent, etc. combining LLMs with deterministic tooling, templates, and validation checks.
• Develop prompt strategies, retrieval and grounding pipelines (e.g., standards libraries, controlled terminology, study specifications), and guardrails for safe, compliant outputs.
• Implement evaluation frameworks for generated code (unit tests, statistical checks, conformance to standard, reproducibility) and establish quality metrics/KPIs.
• Build adapters that integrate agents with code repositories, metadata stores, execution environments, and review/approval workflows; enable provenance and traceability.
• Fine‑tune or customize models where appropriate (e.g., instruction tuning, adapters), and manage model selection, versioning, and inference optimization.
• Collaborate with statistical programmers, data scientist, and statisticians to capture requirements, encode domain logic, and incorporate feedback into agent behavior.
• Partner with full‑stack and DevOps engineers on deployment, monitoring, cost/performance tuning, and reliability in production environments.
• Maintain rigorous documentation of model behavior, data sources, prompt templates, evaluation results, and change control to support audits.

• Master’s degree or above in Computer Science, Data Science, Applied Mathematics, or related field, or equivalent practical experience.
• 3–5 years of experience building AI/ML or NLP applications, including production-grade systems using large language models or sequence models.
• Strong programming skills in Python, with experience in building services and pipelines (e.g., FastAPI, LangChain/LlamaIndex or equivalent frameworks).
• Experience with prompt engineering, retrieval‑augmented generation, and tool/function calling; ability to design deterministic post‑processing and validators.
• Familiarity with software engineering best practices: version control, testing, CI/CD, containerization, and observability for ML applications.
• Experience evaluating generative systems (human‑in‑the‑loop review, rubric design, offline/online metrics, A/B tests) and implementing safety/guardrail mechanisms.
• Ability to translate domain requirements into model capabilities and to communicate tradeoffs among quality, cost, latency, and interpretability.
• Experience integrating LLMs with code generation workflows and execution sandboxes, including static analysis and auto‑testing for generated code.
• Exposure to regulated environments (GxP, CSV) and audit-ready documentation practices; understanding of data privacy and security principles.
• Experience with vector databases, embeddings, and knowledge graph/RAG techniques; model optimization (quantization, distillation) and prompt versioning.
• Familiarity with MLOps for generative AI (model registries, feature/knowledge stores, inference gateways) and cost/performance monitoring.

AstraZeneca