The scope of AI workloads this platform must support is broad. Each Pfizer domain (i.e., R&D, Commercial, Global Supply, Enabling Functions) has distinct compute, latency, governance, and reliability requirements, and this platform must serve all of them without compromise. As Pfizer advances from assistive AI tools toward autonomous agentic systems that take multi-step actions across the enterprise, the demands on this platform will grow in both complexity and consequence. The LLM gateway, agent orchestration layer, and observability infrastructure this leader builds today must be architected for that future from the outset.

The team of engineers is organized across four pods, LLM Gateway & Model Serving, Compute & Environments, Runtime Enablement and Registry, Deploy & Trust, each owning a distinct and critical layer of the AI infrastructure stack including agent lifecycle management.

ROLE RESPONSIBILITIES

Gateway & Serving

• Enterprise LLM gateway, access control, multi-model routing, rate limiting, cost attribution, and audit logging for all LLM interactions across Pfizer, including agentic AI workloads.
• Model serving infrastructure, low-latency inference, auto-scaling, and multi-region deployment for production models.
• Agentic AI runtime, the infrastructure layer that supports autonomous AI agents taking multi-step actions across Pfizer's systems. This is meaningfully different from stateless LLM inference: agents require stateful process management, short-term and long-term memory, tool-calling orchestration, and the ability to coordinate with other agents. As Pfizer's agentic AI portfolio grows, this layer becomes one of the most strategically critical components of the platform. The Head of AI & Agentic Platform Engineering is expected to architect this capability proactively, not wait for agent use cases to arrive and then retrofit the infrastructure.
• Gateway observability, real-time usage monitoring, cost attribution by team and use case, and anomaly detection. Enterprise tool and MCP registry, the governed catalog of tools, APIs, and data sources that AI agents are permitted to call at runtime. As the number of agent-callable tools grows across Pfizer, this registry becomes the mechanism by which the platform enforces what agents can do, not just what they can say. Built and maintained in close partnership with the Trusted AI team's agent governance function.

Compute & Environments

• Enterprise compute provisioning, GPU, TPU, and CPU infrastructure across cloud and on-premises, including capacity planning, FinOps governance, and utilization optimization.
• Pre-configured AI environments, reproducible, governed workspaces that enable data scientists to focus on scientific problems, not infrastructure.
• Infrastructure as Code, automated, auditable environment provisioning across development, staging, and production.
• HPC support, infrastructure capable of supporting large-scale scientific simulation and molecular modeling workloads (preferred, not required).

Runtime Enablement

• MLOps platform, experiment tracking, model versioning, automated evaluation, deployment pipelines, and model registry, with integration into Trusted AI's risk classification and sign-off process.
• Production observability, monitoring, alerting, and dashboarding for AI systems in production: latency, throughput, drift detection, and model health.
• Developer experience, APIs, SDKs, and documentation that enable federated teams to deploy production models without deep infrastructure expertise.

Registry, Deploy & Trust

This pod was previously a standalone Trust Engineering team. Its integration into AI & Agentic Platform Engineering reflects a deliberate architectural decision: agent lifecycle management, register, deploy, monitor, govern, retire, is infrastructure, and the operational boundary between deploying a model and operating an agent has collapsed. The pod owns:

• Enterprise AI model registry, the authoritative record of every AI model and agent in development, staging, and production across Pfizer, including metadata, version history, risk tier, Trusted AI validation status, ownership, and complete audit trail.
• Deployment pipeline infrastructure, automated pipelines through which models and agents move from development to staging to production, with Trusted AI sign-off gates enforced as first-class pipeline steps. Includes release management, canary deployments, A/B testing, and rapid rollback capabilities.
• Production monitoring and drift detection, continuous observation of AI system performance in production: prediction quality, output distributions, latency, throughput, and drift. For agentic systems, monitoring extends to agent behavior, action sequences, tool usage, decision consistency, and anomalous behavior detection.
• Guardrails and policy enforcement, the technical implementation of Trusted AI's governance policies as executable runtime controls: input/output filtering, PII detection, agent action controls, permission scoping, circuit breakers, and prompt injection defenses designed in partnership with the CISO organization.
• GxP-compliant audit trail, complete, tamper-evident logging of every deployment event, configuration change, and model transition, meeting the documentation standards required for AI systems operating in regulated pharmaceutical environments.

• 12+ years in software or infrastructure engineering, with 7+ years in AI/ML platform, MLOps, or AI infrastructure roles at significant scale.
• Demonstrated experience building and operating multi-tenant AI/ML platform infrastructure, compute provisioning, model training pipelines, model serving, and production monitoring.
• Deep hands-on experience with LLM gateway or model serving infrastructure, multi-model routing, inference optimization, access control, and cost attribution at enterprise scale.
• Proven MLOps platform experience with documented outcomes in deployment velocity, reliability, and developer satisfaction.
• Strong IaC practices in a multi-cloud architecture (Azure, AWS, GCP including Terraform expertise.
• Experience leading platform teams with an SLA-driven, product-minded operating model.
• Demonstrated ability to collaborate across organizational boundaries, with adjacent platform teams, security functions, and governance stakeholders.
• Ability to translate infrastructure architecture and trade-offs for both technical teams and senior business stakeholders.
• Experience with encryption and security tools, techniques, and best practices.
• Experience operating AI infrastructure in a regulated environment with GxP controls, audit trail requirements, and validated environment obligations.

• Candidate demonstrates a breadth of diverse leadership experiences and capabilities including: the ability to influence and collaborate with peers, develop and coach others, oversee and guide the work of other colleagues to achieve meaningful outcomes and create business impact.

PRFERRED QUALIFICATIONS

• Experience building or operating ML platform infrastructure at a major technology company (Google, Meta, Microsoft, OpenAI, or equivalent) at petabyte scale with thousands of concurrent ML engineers.
• Experience designing agentic AI infrastructure, specifically the orchestration layer, memory architecture (short-term context, long-term persistent memory), tool-calling and MCP integration, agent-to-agent communication, and the safety architecture required to constrain autonomous agents operating in production. Candidates who have built or operated agent runtimes at scale, whether in a research or product context, will be strongly preferred.
• Deep LLM-specific infrastructure experience: KV cache management, speculative decoding, quantization trade-offs, and concurrent multi-model serving.
• HPC environment experience, job schedulers (SLURM, LSF, or equivalent), parallel file systems, and large-scale scientific compute workloads.

Other Job Details:

• Last Day to Apply: 7/28/2026
• Work Location Assignment: Hybrid