Agentic AI Architect

Agentic AI Architect Role Summary

Design, build, and scale agentic AI systems that plan, reason, and act across complex enterprise workflows. You’ll own the reference architecture, enforce safety and governance, and lead cross-functional delivery to measurable business outcomes. (Microsoft-first environment; multi-cloud familiarity welcome.)

• Architecture &
  
  Roadmap:
  
  Define end-to-end agentic AI reference architecture (planning, tools, memory, orchestration, evaluation, observability, security).
• Use-Case Delivery: Lead discovery → design → production for high-impact workflows (service ops, field ops, finance, claims).
• Agent Orchestration: Implement single/multi-agent patterns with tool use (APIs, RPA), long/short-term memory, and human-in-the-loop.
• Knowledge & Retrieval: Stand up retrieval layers (Azure AI Search/Vector Search, Fabric, pgvector/Pinecone) and optimize grounding.
• Safety & Guardrails: Operationalize policies for PHI/PII, prompt-injection/jailbreak defense, content filtering, red-teaming, and approvals.
• LLMOps/MLOps: Establish eval harnesses, CI/CD, online/offline metrics, rollout strategies, cost/latency/error budgets, drift monitoring.
• Platform Integration: Integrate with Power Platform (Power Automate/Apps/Copilot Studio), Service Now, SAP, queues/events/APIs.
• Migration & Modernization: Convert legacy RPA (e.g., UiPath) to agents or durable cloud workflows where appropriate.
• Standards & Reuse: Publish patterns, accelerators, templates; mentor squads; run design reviews and ensure Definition of Done.
• Stakeholder Leadership: Partner with Security, Legal, Compliance, Data, and business owners; present architecture and results to execs.

• Agentic Stack (Microsoft-first): Azure OpenAI;
  Azure AI Foundry Agent Service / AI Studio;
  Copilot Studio;
  Semantic Kernel/Lang Chain;
  Durable Functions/Logic Apps;
  Event Grid/Service Bus.
• Retrieval & Data: RAG best practices, embeddings, chunking & grounding, vector DBs, knowledge graphs, lineage.
• Engineering: API design, Python/Type Script, async/event-driven systems, containers, IaC (Bicep/Terraform).
• Security & Governance: Entra , secrets management, network isolation, data/model governance, auditability.
• Safety & Guardrails: Threat modeling; prompt-injection/jailbreak defenses; content/PII filters; policy-as-code (e.g., Azure AI Content Safety); red-teaming playbooks; regulated-data handling (HIPAA/PCI/GDPR).
• Agent Evaluations: Offline/online eval design; task-success & tool-use accuracy; function-call correctness; hallucination/harms tracking; golden sets & synthetic tests; judge-model workflows; HITL review; A/B testing.
• Evaluation & Observability: Prompt/test suites, telemetry & tracing, QoS/latency/error budgets, drift/quality monitoring, rollback plans.
• Cost Estimation & Fin Ops: Token-level cost models, per-run unit economics, SLO-aware budgeting, routing/model-selection trade-offs, prompt compression/caching, batch vs. real-time serving, TCO across clouds, alerts & spend guardrails.
• Delivery: Agile leadership, estimation, vendor selection, change management, enablement.

• 7–12+ years in solution/enterprise/AI architecture, incl. 2–4+ years hands-on with LLMs/agent systems.
• Proven track record taking AI from POC → pilot → scaled production in large or regulated enterprises.
• BS/MS in CS/EE or equivalent experience.
• Preferred certs: UiPath Maestro, Azure Solutions Architect / AI Engineer, Power Platform Solution Architect; security/data certs a plus.

Cognizant will only consider applicants for this position who are legally authorized to work in the United States without requiring company sponsorship now or at any time in the future.