AI / Backend Engineer (Regular/Senior)

Soter builds an AI platform for EHS and insurance — production agentic system, real customers, ~15 engineers (5 on AI). You'd be on a small team with high ownership.

We’re looking for an AI/backend engineer who can own significant parts of our AI engine and the backend services that support it. This is a production engineering role — you’ll ship agentic workflows to real customers, hold systems end-to-end, and iterate based on evidence. We care more about high agency and architectural instincts than years on a CV. If you’re genuinely excited about where AI engineering is going and want to work on hard problems with a small, fast team, this is for you.

• You’ll own significant parts of our AI engine: the orchestration layer, LLM tool integrations, and the reliability layer around streaming and structured outputs
• You’ll architect and ship agentic workflows end-to-end — agent boundaries, tool interfaces, failure handling, and human oversight points
• You’ll drive AI quality: define success criteria before shipping, build and run eval sets, catch regressions before users do, iterate on evidence not gut feel
• You’ll own AI production operations: trace LLM calls and agent steps across the stack, monitor cost and latency, respond to incidents
• You’ll hold backend services end-to-end across our Python microservices — schema, API, deploy, on-call
• You’ll keep AWS infra (Terraform, Ansible) and CI/CD boring and reliable
• You’ll raise the engineering bar — clean code, the testing pyramid, sharp code reviews

• Production experience building LLM-powered solutions — agents, tool calls, prompt pipelines — not just using AI tools or experimenting on pet projects
• Hands-on context architecture: prompt engineering, structured outputs, schema validation, few-shot design, context window management
• Experience building and operating agentic systems: tool interface design, orchestration patterns, failure handling, agent state management
• Systematic approach to AI quality: eval sets, success criteria, failure pattern analysis, evidence-based iteration
• Proficiency in Python (production-grade, enterprise experience)
• Solid backend fundamentals: APIs, microservices, SQL database design and optimisation
• Strong architectural and design instincts — you can reason about system design clearly, verbally and visually
• Demonstrated ability to work autonomously and own systems end-to-end
• Daily hands-on use of AI development tools (Cursor, Claude Code, Copilot, or similar) — this is a hard requirement; we care about how you use the tool, not which one
• Fluent English (written and verbal)
• Self-driven, product-minded, high agency — no hand-holding needed

• Has owned a non-trivial AI feature or production service end-to-end for 12+ months — design, deployment, on-call, iteration on real user feedback

• Build and ship a new agentic workflow end-to-end — design, tools, evals, rollout to a real client
• Tackle a class of LLM reliability issues (e.g. streaming timeouts with reasoning models, gateway fallback edge cases)
• Close observability gaps so a single conversation can be traced cleanly across our stack

• Experience with LLM orchestration frameworks (LangChain, LlamaIndex, LangGraph, etc.)
• Multi-agent system design and operation
• Model routing, cost governance, or LLMOps tooling
• Familiarity with evaluation frameworks (LangSmith, RAGAS, custom harnesses)
• Observability tooling (Datadog, Grafana, OpenTelemetry, Langfuse)
• AWS infrastructure experience (Terraform, Ansible)
• Node.js or TypeScript backend experience

• Join a small team of passionate engineers dedicated to innovation and excellence
• Work on a product that genuinely improves people's lives and workplace safety
• Experience a startup culture: fast-paced, close collaboration, real influence on key decisions
• Short feedback loops — ship fast, learn fast
• Minimal bureaucracy — focus on what matters: building great software
• AI-first engineering culture — we embrace and invest in AI-augmented development