What the work involves

You own the delivery path for LLM applications running on Azure. Day to day that means writing and maintaining YAML pipelines in Azure DevOps, provisioning infrastructure through Terraform, Bicep, or ARM, and shipping containerised workloads to AKS with images in Azure Container Registry. The AI-specific part is real: LangGraph agent workflows have stateful checkpointing, long-running executions, per-node model calls, and failure modes that don't look like ordinary web services, so you will be designing deployment topology, secret handling through Key Vault, and observability in Azure Monitor and Application Insights with those behaviours in mind. Expect to sit between the AI/ML team and platform engineering, absorbing questions about latency, token cost, and why an agent hung in production.

What the screen looks for

Turing's screening is AI-led and follow-up heavy. It will take a claim — "I deployed LangGraph agents to AKS" — and press on specifics: which checkpointer backend, how you handled concurrency, what the pipeline stages were, what broke. Vague platform-level fluency is detected quickly. The strongest signal you can give is a concrete architecture you personally built, described with service names, versions, and the tradeoff you chose against an alternative. Because this is an AI-application role rather than pure infrastructure, expect at least one probe on how you evaluate or monitor non-deterministic behaviour — regression on agent outputs, not just uptime.

Logistics

  • Remote, based in India, working with distributed development and AI/ML teams.
  • Availability immediate to two weeks; late start dates are a common reason candidates are deprioritised.
  • Agile/Scrum cadence, with overlap expected for standups and incident response.
  • Pay band undisclosed by the platform for this listing. Turing typically discusses rate after the technical screen; nothing here is guaranteed.

Azure certifications (AZ-104, AZ-400, AZ-305) and exposure to LangChain, Azure OpenAI Service, RAG pipelines, or vector databases are preferred rather than required, but they shorten the conversation.