Data & AI Platforms Lead, BAIC

About the Beijing AI Center The Beijing AI Center is a new strategic investment by AstraZeneca to accelerate drug discovery through AI. The center brings together AI researchers, computational scientists, and platform engineers to apply foundation models, agentic AI, and large-scale scientific computing to real R&D problems. Situated in one of the world’s most dynamic AI talent markets, it operates at the intersection of AI and biologics discovery, computational chemistry, and data-driven drug development. The center is structured around three pillars: Discovery verticals (biologics engineering, computational chemistry) that own the science; Data & AI Platforms (this role) that own the capabilities; and R&D IT that owns the infrastructure. A dedicated on-premises GPU cluster provides the compute backbone, operated by IT and shaped by platform standards. About the Role Help build AstraZeneca's Beijing AI Center from the ground up, and own how AI gets done there. This is the person who turns a new on-premises GPU cluster, a fast-growing team, and China's foundation-model ecosystem into platform capabilities that make drug-discovery science faster. You set the methods, standards, and tooling that sit between raw infrastructure and the scientists using it. Success is the adoption, scale, and reuse of those capabilities across Discovery teams, not the delivery of any single AI project. The center runs on three teams that depend on each other: Discovery verticals (biologics engineering, computational chemistry) own the science; Data & AI Platforms - this role - owns the capabilities; and R&D IT owns the infrastructure. A dedicated on-premises GPU cluster provides the compute backbone, operated by IT and shaped by your standards. As Enterprise AI's single point of accountability for the center, you are who global and local stakeholders come to when a capability needs standing up, a bottleneck removed, or an ad hoc problem solved. You lead through your team and through the product owners you direct rather than building everything yourself; what you bring personally is enough technical depth to set the bar and judge the work. You own the requirements, methods, tooling, and evaluation rigor that make Discovery and IT more productive. IT handles GPU provisioning, cluster operations, and networking; Discovery owns model architecture, training objectives, and scientific interpretation. On the shape of the ideal candidate: this is a broad mandate, and we are not looking for equal mastery of all five areas below. We expect deep strength in two or three of them and credible command of the rest, with the judgment to lead the others through strong specialists. What You Will Do Five focus areas, each roughly a fifth of the mandate. Priorities will shift as the center scales. 1. Platform Leadership - Single Point of Accountability Be the single point of accountability for the Beijing AI center: triage needs, remove bottlenecks, and problem-solve across teams so it succeeds. Serve as the escalation and decision point for platform, data, and compute needs across the center, owning the resolution, not just routing it Proactively clear blockers spanning IT, Discovery, global platform teams, and external partners Run the standing cross-functional coordination where center priorities and trade-offs are decided Translate ambiguous, fast-moving priorities into a coherent platform delivery plan Represent platform capabilities to global AI leadership and ecosystem partners Mandate: you are accountable for center-level platform outcomes, not just your team's deliverables. Where a need falls between teams, you own resolving it - usually through influence rather than formal authority. 2. Compute Enablement & Research-Computing Strategy Own the demand side of the compute interface with IT, and the engineering strategy that lands research workloads on shared infrastructure efficiently. Own the compute demand-and-supply picture: gather workload requirements, forecast demand, and map scenarios IT can plan and procure against Set the research-computing engineering strategy: how workloads are packaged, optimised, and served including inference and serving strategy (quantisation, batching, throughput) for the models the platform relies on as well as the scheduling and orchestration requirements handed to IT Define and enforce MLOps standards - experiment tracking, model registry, CI/CD for ML, environment and dependency management - so compute is productive out of the box Set the direction for a sound data foundation for AI (AI-ready pipelines, harmonisation, retrieval and embedding infrastructure for scientific and literature data), working with IT and the data organisation who own the data platform Boundary with IT: IT operates the cluster - provisioning, hardware, networking, scheduling execution, vendors. You own the requirements, forecasting, MLOps standards, and the engineering approach that makes it productive for research. 3. Agentic AI Platform - Direction & Delivery Set the direction for the center's agentic AI platform and deliver it through a product owner and their team. Set the vision and roadmap: multi-agent orchestration, local agent development frameworks, and scientific workflow automation Direct delivery through a product owner and team - setting outcomes, holding delivery to account, and unblocking, rather than building it personally Deliver against measured impact: adoption, developer productivity, scientific enablement, and business value Judge what to build, buy, or partner for: evaluate the China LLM landscape (Qwen, DeepSeek, GLM, Moonshot) and assess local partnerships (e.g., Alibaba, Tencent, Baidu) for inference, agent hosting, and platform capabilities Set the methodology for evaluating agent reliability, safety, and scientific accuracy, and define - with the product owner and IT - who owns runtime guardrails and tool-use permissioning once agents touch sensitive systems Boundary: IT provides hosting infrastructure; a product owner runs the hands-on build. You own the direction - what gets built and why, which LLMs and partners are selected, and how impact is measured. 4. AI Engineering - Standards, Evaluation & Depth Set the center's AI engineering bar: the methods, standards, and evaluation frameworks that make the work reproducible and comparable. You hold enough command of the methods to set the standard and judge the work - the depth is in the judgment, evidenced by a track record of the calls made, not in running every job yourself. Set the standards for fine-tuning, post-training, and inference optimisation of open-weight models - method selection, training configurations, reproducibility, checkpoint and seed management, result validation - and judge the team's work against them Direct the center's evaluation platform: test harnesses, metric dashboards, leaderboards, and evaluation-as-a-service - including the technical assessment behind which models to adopt, and evaluation of agent behaviour and tool-use reliability Curate domain-relevant benchmark suites with Discovery (biologics, computational chemistry): metrics, held-out test construction, and data-leakage controls Establish objective criteria to compare models, tools, and approaches, enabling evidence-based technology decisions Boundary with Discovery: Discovery scientists select model architectures, define objectives, curate domain dat