Senior Software Engineer - AI Engineering
Mercury · Portland, OR · 1 wk ago
Engineering$167k–$219k/yrFull-time
What you'll do
- Build and evolve MCP servers that connect internal systems and data sources into a coherent interface for agents and engineers.
- Expand and operate our LLM gateway infrastructure: routing, rate limiting, cost attribution, and observability across teams.
- Turn early patterns into durable defaults: shared prompt libraries, guardrails, and policy-as-code so teams can move fast safely.
- Shape and maintain structured context artifacts—clean, reliable, agent-consumable—so LLMs working in Mercury's systems can reason accurately about our domain.
- Improve internal knowledge discoverability and retrieval so both humans and agents can quickly find accurate answers.
- Partner with domain teams to standardize key sources of truth, and keep them fresh.
- Build and refine sandbox environments and tooling that let engineers experiment with AI safely and at speed.
- Create self-service scaffolding so non-engineers—PMs, ops, finance—can prototype and deploy AI-powered workflows with minimal hand-holding.
- Build playgrounds and evaluation harnesses so internal AI agents can be tested and iterated in controlled environments before hitting production.
The ideal candidate
- Has 5+ years of backend development experience in complex, production systems—you've built things that other engineers depended on.
- Is fluent across programming languages and can navigate platform engineering, infrastructure, and developer tooling without needing a map.
- Has hands-on experience building LLM-powered systems—RAG pipelines, agents, eval frameworks—and has shipped at least one of these to production.
- Understands the real tradeoffs in AI deployments: cost modeling, observability, latency, and safety—not just the exciting parts.
- Is high-agency and self-directed. You can operate effectively without tightly-defined scope, find the highest-leverage work, and get it done.
- Communicates clearly across technical and non-technical audiences—you can explain what you built and why it matters.