AI & analytics engineering (~60%)

• Own end-to-end build of insight pipelines on Databricks: certified Delta tables → metric layer → signal detection → LLM narrative → application surface.
• Design agentic workflows on an agent SDK (Claude Agent SDK or equivalent): tool design, context and prompt engineering, governed context stores, fallbacks and failure handling.
• Enforce guardrails in data and metadata — governed tables, SQL-level scope control — rather than in prompt instructions.
• Build evaluation harnesses for LLM output: answer-quality rubrics, binary gates (numerical fidelity, grounding, compliance), regression suites, release gates; tracing via Langfuse/LangSmith or equivalent.
• Deliver conversational analytics (NL-to-SQL) on governed semantic layers (Databricks Genie, Metric Views) with certified KPI definitions.
• Set engineering standards: code review, Git/CI, unit and SIT test plans, UAT support, release notes, observability, cost and latency tracking.
• Produce architecture documents, security / IT-risk evidence packs and status reporting with honest status legends (implemented / in-flight / planned / blocked).

Core commercial analytics (~40%)

• Take ambiguous questions from any client function — commercial, medical, market access, clinical development, supply chain, enterprise operations — and structure them into decision-oriented analytical plans; own them end to end.
• Pick the method the question needs — diagnostic analysis, cohort and longitudinal analysis, segmentation, forecasting, test-and-control and causal inference, predictive modelling, optimisation — and defend the choice.
• Work across the client's data estate — syndicated and patient-level real-world data, clinical and trial data, engagement and CRM data, payer and access data, operational and financial data — integrating sources and resolving definitional conflicts.
• Establish governed definitions (KPIs, cohorts, business rules) with the client and make them enforceable in code — one definition, every surface.
• Convert results into decisions: QC'd, client-ready narratives and decks; activation into workflows, dashboards or products; present to Director+ stakeholders.

Client & team leadership

• Plan, review and coach a team of 2–4; QC everything before it reaches the client.
• Own workstream status, risks and dependencies with ZS leadership and client PMO.
  • Contribute to proposals, POVs and CST capability building — reusable assets and AI upskilling of CST talent.