Senior Director, AI Architecture

Royal Bank of Canada; Toronto; Sur place; CAD 180 000 - 280 000; Plein temps

Résumé du poste

Royal Bank of Canada is seeking a senior AI architecture leader to standardize AI across technology and operations. You will own the reference architecture for RAG, AI agents, model access and AI-native software delivery, driving patterns that scale and pass audits.

You will champion secure sandboxing, governance, and enterprise-wide quality gates. The role requires deep expertise in LLM systems, security, DevSecOps, and OSFI-aligned controls, with a track record of delivering at enterprise

Qualifications

• 15+ years in architecture or engineering with enterprise-scale leadership.
• Hands-on depth in LLM systems: RAG, agent frameworks, evaluation and inference serving.

Responsabilités

• Define enterprise AI reference architecture and patterns as code.
• Unify RAG subsystems into a single platform for ingestion, embeddings and retrieval.
• Design agent sandboxing, secure access, and credential management.
• Own AI harness architecture with deterministic quality gates (tests, static analysis, dependencies).
• Manage evaluation lifecycle and re-run evals before model upgrades.
• Architect model gateway routing by cost, risk and jurisdiction; enable metering and caching.

Connaissances

LLM systems

RAG platforms

Agent frameworks

MCP

Evaluation & inference

Security architecture

DevSecOps

SDLC

Executive communication

Description du poste

Job Description

What is the opportunity?

This leader sets one standard way to build, secure and run AI across RBC Technology & Operations. They own the reference architecture for retrieval-augmented generation (RAG), AI agents, model access and AI-native software delivery (AI-DLC). The role turns the 2030 T&O ambitions, including AI-native delivery and 30% of GenAI volume on open-weight models, into patterns teams actually use. Success means fewer bespoke stacks, faster delivery with maintainable code, and AI that Internal Audit, Model Risk and OSFI can trust.

What will you do?

AI architecture standards as code. Define the enterprise AI reference architecture and approved patterns. Encode them as harness skills, agent constitutions and automated fitness checks, so standards apply at generation time rather than at review.

Common RAG substrate. Converge the many RAG solutions onto one shared platform: ingestion, chunking, embeddings, hybrid retrieval, reranking and evaluation. Retrieval must respect source entitlements by default. Set a retirement plan for duplicate stacks.

Agent sandboxing and security architecture. Design the sandbox standard for every agentic workload: ephemeral environments, egress allow-lists, short-lived task-scoped credentials and no standing production access. Own agent identity, tool access through a governed MCP gateway and registry, and defenses against prompt injection and poisoned tools.

AI-DLC harness and quality gates. Own the harness architecture from spec to release. Make quality gates deterministic and mandatory: tests and coverage, static analysis, complexity and duplication limits, dependency and license checks, security scanning and architecture fitness checks. Enforce separation of duties (the agent that writes a change cannot approve it), PR size limits and traceability from spec to deployment.

Skills and evaluation lifecycle. Treat skills like code: owned, versioned, regression-tested with evals and retired when stale. Re-run evals before any model upgrade reaches teams.

Model gateway and inference architecture. Route work across frontier and open-weight models by cost, risk and jurisdiction. Build in token metering, caching and data-classification rules so cost and residency are enforced by the platform.

Risk-tiered control model. Partner with Internal Audit, Model Risk and Cyber to define human-oversight tiers, from human-on-the-loop for low-risk work to human-in-the-loop for payments and core banking. Align with OSFI Guidelines E-23 and B-13.

Design authority and innovation. Chair AI architecture review, keep decision records and manage exceptions. Graduate proven lab work into the platform and evaluate emerging technology, such as agent protocols and new open-weight models.

What do you need to succeed?

15+ years in architecture or engineering, including 5+ years leading architects or engineers at enterprise scale. Hands-on depth in LLM systems: RAG, agent frameworks, MCP, evaluation and inference serving on cloud and on-premises GPUs. Security architecture for AI, including the OWASP Top 10 for LLM applications, sandboxing and identity for non-human actors. Strong DevSecOps and SDLC background; you have shipped quality gates engineers accepted rather than routed around. Experience in a regulated industry, ideally financial services under OSFI. Credibility with engineers: you still prototype and read code. Able to explain trade-offs plainly to executives and audit.

What's in it for you?

We thrive on the challenge to be our best, progressive thinking to keep growing, and working together to deliver trusted advice to help our clients thrive and communities prosper. We care about each other, reaching our potential, making a difference to our communities, and achieving success that is mutual.

A comprehensive Total Rewards Program including bonuses and flexible benefits, competitive compensation, commissions, and stock where applicable

• Leaders who support your development through coaching and managing opportunities
• Ability to make a difference and lasting impact
• Work in a dynamic, collaborative, progressive, and high-performing team
• A world-class training program in financial services
• Flexible work/life balance options
• Opportunities to do challenging work

#LI-POST

#TECHPJ

Job Skills

• Big Data Management
• Data Modeling
• Data Science
• Decision Making
• Deep Learning
• Machine Learning (ML)
• Predictive Analytics
• Programming Languages
• Relationship Building

Additional Job Details

• Address:
• RBC WATERPARK PLACE, 88 QUEENS QUAY W:TORONTO
• City: Toronto
• Country: Canada
• Work hours/week: 37.5
• Employment Type: Full time
• Platform: TECHNOLOGY AND OPERATIONS
• Job Type: Regular
• Pay Type: Salaried
• Posted Date: 2026-10-27
• Application Deadline: 2026-11-20
• Note : Applications will be accepted until 11:59 PM on the day prior to the application deadline date above

Our Employment Opportunities

At RBC, we are guided by living shared values of Client First, Integrity, Collabor