Site Reliability Engineer (AI Observability)

Appnovation; Montreal (administrative region); On-site; CAD 90,000 - 135,000; Full time

Job summary

Appnovation is seeking a Site Reliability Engineer to build a shared observability platform for LLM-based applications. You will own monitoring, alerting, and SLAs from design through implementation, covering ClickHouse, PostgreSQL, Redis, and the Kubernetes layer.

The role emphasizes infrastructure as code and runbooks that guide on-call teams during incidents. You’ll work with Langfuse, Argo CD, and a self-hosted Kubernetes stack on AWS, ensuring robust backup/restore, upgrade strategies, and

Qualifications

• Experience with Kubernetes on AWS (EKS) and Helm values.
• Experience running ClickHouse in production including replication and Keeper quorum.
• Ability to design monitoring, alerting and service levels reflecting user experience.

Responsibilities

• Build monitoring, alerting and service levels from scratch.
• Diagnose and resolve failures across ClickHouse, PostgreSQL, Redis and the Kubernetes layer.
• Own backups, restores, and rehearsals with rollback plans for migrations.
• Create runbooks and SOPs enabling on-call staff to resolve incidents independently.

Skills

Observability engineering

Incident management

Automation

Runbooks writing

Client-facing collaboration

Tools

Kubernetes

Langfuse

Argo CD

ClickHouse

PostgreSQL

Redis

Helm

Job description

• Appnovation is a global, full-service digital partner that combines Strategy, Experience & Design, Engineering and Managed Services. We build digital solutions that deliver real impact today and serve as foundations for future growth. Bold ambition. Practical action. Endless possibilities
• We’re looking for a Site Reliability Engineer to keep a shared observability platform for LLM-based applications running for a global life sciences client
• The platform is built on Langfuse and self-hosted on Kubernetes on AWS, with ClickHouse as the analytical store, PostgreSQL for metadata and Redis as the ingestion queue, all delivered through Argo CD
• Two things need building rather than maintaining. The platform has no monitoring, alerting or defined service levels today, and you will own putting them in place. Infrastructure is defined as code throughout, in Kubernetes manifests, Helm values and Argo CD applications
• Alongside the platform itself, you will own the runbooks that make an incident survivable by someone other than the author, and the onboarding and support path for the internal teams that depend on the platform
• Platform Operations: Diagnose and resolve failures across ClickHouse, PostgreSQL, Redis, the ingestion workers and the Kubernetes layer beneath them, including ingest backpressure from queue depth and worker drain behaviour
• ClickHouse Operations: Own ClickHouse under the operator model, including Keeper quorum, replication, shard and replica topology, and S3 storage tiering
• Provable Backup and Restore: Rehearse the restore, time it, document it and test it against its failure modes, rather than assuming a successful backup job means a recoverable system
• Safe Upgrades: Plan and rehearse upgrades in a lower environment, with a rollback plan that works even if a migration is only partly complete
• Monitoring and Service Levels: Build monitoring, alerting and service levels from scratch, so problems are found here before a user reports them
• Infrastructure as Code: Maintain Kubernetes manifests, Helm values and Argo CD applications so every change goes through the delivery pipeline
• Runbooks and SOPs: Write and maintain runbooks and SOPs that let a colleague resolve an incident without the author present
• Onboarding and Support: Run the onboarding and support path for internal teams that depend on the platform, and triage what they bring
• Automation: Turn recurring operational work into automation
• Upgrade Partnership: Work with the platform engineer who owns what the platform offers. They decide what to adopt and how it is configured; you own the migration and its rollback

Experience upgrading self-hosted software safely, including schema migrations, rehearsal in a lower environment and a rollback plan for a partly completed migrationHands-on experience with Kubernetes on AWS (managed EKS), with routine work done through Helm values and Argo CD applicationsHands-on experience running ClickHouse in production, including replication and Keeper quorum, shard and replica topology, and backup and restore, ideally run through an operatorExperience building monitoring and alerting from scratch, including service levels that reflect what users actually experience rather than what is easy to measurePostgreSQL and Redis operations deep enough to debug metadata-store and queue problems, including backpressure and worker drainStrong operational writing: runbooks, SOPs and post-incident reviews that a colleague can follow unaided during an incidentYou want to find problems before users doYou write runbooks for the person on call at 3am, not for yourselfYou don’t trust a backup until you’ve restored from itYou stay calm in incidents and focus on fixing the process afterwardYou work well inside a client team and build trust quicklyYou have prior experience in consultingYou automate anything you have to do twicePrior experience and connections in the Life Sciences industry is preferredObservability engineering, including OpenTelemetry Collector pipelines, alerting design and Grafana dashboardsOIDC or enterprise SSO integration with a corporate identity providerGitHub Actions for plan and apply pipelines with approval gatesExperience running LLM observability tools such as Langfuse, LangSmith or Arize PhoenixExperience in pharma, life sciences or another regulated industry