The opportunity
Coursera’s AI adoption has moved fast — engineering teams code AI-first, dozens of internal tools have shipped, and functions across the business are building their own AI agents and automations. The next phase requires something different: a shared technical foundation that…
What you'll do
Context & Semantic Architecture: Design how AI agents access and understand Coursera’s internal data and knowledge, so agents retrieve the right information rather than guessing or over-fetching. Establish reusable patterns for RAG, semantic search, structured data access, and enterprise context, partnering with the Data team on existing data models and pipelines.
Agentic Architecture & Harness Engineering: Design reusable patterns for agent orchestration, tool use, multi-step workflows, state management, and human-in-the-loop decisions. Build the engineering harness around agents, including testing, evaluation, guardrails, retries, observability, and failure recovery, so agentic solutions are reliable and production-ready.
Governance Architecture: Build technical guardrails that enforce—not just document—who can build what, what data and systems an agent can access, and what gets logged for audit and review. Establish standards for agent ownership, lifecycle management, and production readiness.
Security Architecture: Design defenses specific to agentic AI, including agent identity and access control, least-privilege permissions, prompt injection, data leakage, secure credentials, and safe tool execution. Coordinate with Security to extend existing enterprise security standards.
Self-Serve AI Development (“Vibe Coding”) with CI/CD Automation: Extend and mature Coursera’s self-serve deployment path so AI-assisted, conversational development ships through automated testing, security checks, and evaluation gates—safe by default, not safe by manual gatekeeping.
LLM Cost & Performance Optimization: Design caching, model routing (matching task complexity to the right-sized model), token and context optimization, and usage monitoring so AI adoption can scale without costs growing linearly. Establish cost, latency, quality, and usage metrics.
What they're looking for
- + years of experience with clear progression from engineer to senior/staff: engineer to architect leader, including ownership of platforms or infrastructure used by multiple teams—not just individual feature delivery.
- Hands-on experience with agentic AI frameworks and orchestration (e.g.,: LangGraph or comparable) and familiarity with Model Context Protocol (MCP) or similar standards for connecting AI systems to enterprise data and tools.
- Experience designing agent harnesses, evaluation frameworks, guardrails,: testing, observability, and failure recovery for production AI systems.
- Practical experience with enterprise context architecture, RAG, semantic search, or structured data access.
- Practical experience with LLM cost and performance engineering—prompt: caching, model routing, token economics—with quantifiable results.
- AI security fundamentals, including access control, data isolation, prompt: injection defense, and secure tool execution.
- CI/CD, cloud infrastructure, APIs, and production operations experience.
- Experience defining enterprise AI architecture, reusable platform patterns,: operating models, or governance standards across multiple teams.
