Job Title: Senior AI Platform Engineer, Agentic Platform
Location: Montreal, QC (Hybrid)
Role summary:
Build and run the agentic platform behind the firm's model inventory and validation workflows. Agents run in Azure but depend on on-prem authorization, entitlement and storage services, so you will design how the two halves work together and take the result into production under a regulated bank's controls. This is a production engineering role, not a prototyping role.
What you will do:
- Design and build a shared agent runtime: orchestration, state, tool access, retries and timeouts, human approval steps and kill switches, reusable across validation workflows.
- Expose MCS capabilities to agents as governed tools (APIs or MCP servers) over the Artifact Manager APIs.
- Design the hybrid path between Azure-hosted agents and on-prem model risk, entitlement and storage services: connectivity, identity, latency budgets and failure isolation.
- Ensure every agent acts with the requesting user's entitlements across data-owner, reviewer and read-only auditor roles, with a full audit trail.
- Move documents reliably between on-prem NAS, MongoDB Atlas metadata and ADLS Gen2.
- Build services and Kafka consumers on the Model Bus and deploy them to AKS through Jenkins.
- Deliver agents for validation package gating, semi-automated validation and report drafting.
- Own release, monitoring and incident response for what you ship: blue/green releases, runbooks, OpenTelemetry tracing.
Must have:
- 8+ years of software engineering, including 3+ years building and operating production platforms at a bank or another regulated financial institution.
- A system you took through the bank's full path to production: change management, security review, entitlement onboarding and post-go-live support.
- Designed or operated a hybrid architecture where cloud services depend on on-prem systems: private connectivity (such as ExpressRoute), DNS, firewalls, identity federation, and behaviour when the link degrades.
- Built an agent or LLM platform in production, not just an app: tool calling, orchestration, state, guardrails, structured outputs.
- Hands-on enterprise identity: OAuth 2.0 and OIDC, service-to-service authentication, token exchange or on-behalf-of flows, integration with central entitlement systems.
- Strong Python backend development (FastAPI or Flask) and REST API design for enterprise scale; comfortable reading Java and other services you did not write.
- Kubernetes (AKS or OpenShift), Docker and Helm workloads you deployed and operated; Jenkins CI/CD and GitOps.
- Kafka producers and consumers in production, including ordering, idempotency and replay.
Strong plus:
- LangGraph, Semantic Kernel or similar orchestration; MCP and agent-to-agent patterns.
- Model risk management domain: SR 11-7 or OSFI E-23, three lines of defence, model validation.
- Document AI: PDF parsing, table and figure extraction.
- Working familiarity with Databricks or Snowflake as downstream services the agents call.