AI ENGINEER

JOB TITLE: MEDIOR AI ENGINEER — LLMOPS & RUN FOR AI AGENTS AND RAG (AZURE / KUBERNETES)

LOCATION: BRUSSEL BELGIUM

LANGUAGES REQUIRED: DUTCH,FRENCH,ENGLISH

CONTRACT DURATION:04/01/2027 - 31/12/2027

WORK MODE: HYBRID MODE(FULL TIME 2- 3 DAYS ONSITE PER WEEK)

JOB DESCRIPTION

Context

• Client (Brussels public transport operator) is building several employees- and customer-facing AI services, serving our customers and employees across the organization. These range from retrieval-based assistants to AI agents that call tools and business systems on behalf of users. The AI services are built and run on our STIB AI Platform. The first services go live from December 2026, and more will follow in 2027.

Your Roles and Responsibilities

• We are recruiting a Medior AI Engineer to join the AI Centre of Excellence (Digital Innovating & AI - delivery team) and own the Run phase of our AI services in production. You take over each service from the Senior AI Engineers who built it, and you keep it reliable, measured, secure and cost-efficient over time. You will work alongside STIB’s internal AI Platform Architect, two Senior AI Engineers (Build), the business and content owners of each service, and client's IT operations teams.
• The AI Platform runs on a hybrid operating model: managed Azure services (Microsoft Foundry with the OpenAI model family for LLM and embeddings, Azure AI Search, Azure API Management, Azure AI Content Safety) combined with self-hosted open-source components on AKS (agent orchestration in Python with Lang Chain, ingestion pipeline), with MLflow on Databricks for evaluation and tracing (Open Telemetry). The Digital Innovation / AI CoE team owns Infrastructure-as-Code for all application-scoped Azure resources, using Terraform and Azure DevOps pipelines.

What you will deliver

Activities will vary from service to service. Typically, you will:

• operate the AI services in production: monitor availability, latency and errors, handle incidents and service requests, analyse root causes, then fix or escalate
• keep answer quality under control: run and maintain evaluation sets (golden sets) with business content owners, track answer quality, non-answer rate and user feedback, and detect and investigate quality regressions
• operate the ingestion pipelines: incremental refresh, failed documents, index health, and onboarding of new content sources or business domains within existing connectors
• operate agent integrations: monitor tool and integration calls to business systems, handle failures and timeouts, and review action logs and exceptions for agents that act on users’ behalf
• manage change safely: model version upgrades and retirements, prompt and configuration changes, dependency and security updates, each validated against the evaluation sets before release
• follow run cost: token consumption and cost per service, alerting on drift and proposing optimisations maintain the Terraform, Helm and Azure DevOps artefacts, and keep the DEV, TST, ACC and PROD environments consistent
• apply data protection rules: trace retention, erasure and access restrictions, as defined in each service’s data protection impact assessment
• maintain runbooks and documentation, and report service health to service owners and the AI CoE on a regular basis deliver small enhancements and fixes; structural changes and new features stay with the Build team.

SKILLS

• Azure DevOps
• Azure Kubernetes Service - AKS
• Azure OpenAI Service
• Lang Chain
• LLM / Large Language Models
• Python
• RAG (Retrieval-Augmented Generation)
• Terraform
• Vector databases

Questions

• Have you personally operated or supported an LLM application (RAG, agent or document processing) in production for at least 6 months?
• (if answer to Q1 is “Yes”) Please describe the application, its user volume, what you monitored, and your role in incidents and changes. Maximum 200 words.
• Have you personally investigated a quality issue or incident in an LLM application or AI agent using traces or logs?
• (if answer to Q2 is “Yes”) Please describe one case: the symptom, how it was detected, the root cause, the fix, and what you changed to prevent it from happening again. Maximum 200 words.
• Have you run or maintained evaluations of an LLM application, for example re-running test sets after a model or prompt change?
• (if answer to Q3 is “Yes”) Please describe the metrics, how the test sets were maintained, and how the results influenced release decisions. Maximum 200 words.
• Have you personally deployed changes to Kubernetes workloads (manifests or Helm charts) and Terraform code through CI/CD pipelines on Azure?
• (if answer to Q4 is “Yes”) Please give one concrete example. Maximum 150 words.
• This role requires on-site presence in Brussels 2 to 3 days per week. Are you willing and able to commit to this schedule?