What is The Role

The Context Engine team builds the knowledge layer that AI agents use to work with enterprise data in Elasticsearch. We extract knowledge from any data sources into a structured AI Index, serve it to agents through public APIs, MCP tools and framework integrations, and close the loop with agent traces so that what the engine knows improves from real usage. Any agent can use it: Elastic’s own Agent Builder, Claude Code, LangChain and other third-party harnesses.

As a Principal AI Engineer, you own the improvement loop of this product end to end: how agents, automations and skills behave in production, how we observe them, how we evaluate them, and how we ship changes to them safely. This is a hybrid role at the intersection of engineering, data science, and product. You will write production code, design evaluation and telemetry that product decisions can rest on, and set the technical bar for how the team iterates on agentic behaviour. You will work alongside data scientists, backend engineers, product, and UX, and your work will show up directly in what customers build on top of Elastic.

The codebase is TypeScript and we build it in the open, so you'll be shipping code, designs and discussions in public alongside the rest of the Elastic Stack.

What You Will Be Doing

• Own the production improvement loop for Context Engine: understand how extraction automations, retrieval tools and memory behave, based on offline evaluations and customer conversations and telemetry. You help find the failure modes, fix them, and prove the fix.
• Define how we iterate on agents and skills safely: versioning and rollout of prompts, skills and automations, regression coverage, staged and shadow evaluation, and the guardrails that let us change behaviour without breaking customers.
• Design the telemetry we need to make data-informed engineering decisions: what to capture from agent traces, tool calls and knowledge retrieval, how it lands in Elasticsearch, and how it feeds evaluation, dashboards and the feedback loop.
• Partner with the data science team on evaluation strategy: golden datasets, evaluators to gate on quality, latency and cost.
• Raise the bar across the team: review designs and PRs, mentor engineers in eval-driven development, and write the technical proposals that shape the roadmap.

What You Bring

• 10+ years of software engineering experience, with the recent years spent shipping and operating AI-driven products on real production traffic, ideally products with public APIs and data models that had to evolve without breaking customers.
• A track record of eval-driven product improvement: you have diagnosed agent or LLM behaviour from traces and user feedback, designed the evaluation that exposed the problem, shipped the fix and measured the outcome.
• Direct experience building agents with state and memory, and iterating on prompts, skills and tool behaviour safely in production.
• Familiarity with MCP, including exposing public MCP servers and tools.
• Experience designing telemetry for AI systems, and using it to make engineering and product decisions.
• Experience running product experiments end to end: instrumentation, unattended execution, and interpreting results.
• Experience building products with public APIs and evolving data models, and the judgement that comes with maintaining external contracts as a product changes.
• Strong backend engineering skills in either Python or TypeScript: APIs, stateful workflows, data pipelines and production services.
• Comfort working with data scientists, engineers, and product managers as peers, translating between measurement and shipping, and communicating trade-offs clearly.
• A pragmatic, low-ego style suited to a distributed, async-first team.

Bonus Points

• Experience with agent frameworks such as LangGraph, CrewAI, Claude Agent SDK, or similar.
• Agentic retrieval experience, including knowledge representation.
• Practical Elasticsearch experience.