Software Engineer, AI Systems (Canada)
Salary not listed • Canada • Full-time
Posted 1mo ago
About the job
Haven Safety AI is building an enterprise learning intelligence layer for safety, helping high-risk organizations learn faster from incidents to prevent future occurrences. The platform integrates with existing EHS systems to enhance how incidents are investigated, assessed, and learned from. It combines incident evidence, company knowledge, historical cases, and an industry knowledge graph, with specialized AI agents analyzing various factors to produce traceable assessments. This role focuses on the AI and LLM engineering layer, connecting the evidence base and knowledge graph to the product experiences used by investigators and safety leaders. It's a build-and-operate role requiring technical expertise and operational responsibility in a safety-critical, multi-tenant enterprise environment.
Responsibilities
- Build and operate multi-step LLM pipelines coordinating model calls, tool calls, graph queries, retrieval, quality gates, and agent handoffs.
- Extend the coordinated agent team and orchestration layer for incident analysis and enterprise learning.
- Design the context layer using graph traversal, vector search, and hybrid retrieval for accurate model inputs.
- Build datasets, scoring, regression suites, model comparisons, and human-label loops for evaluation.
- Implement production AI operations including tracing, tool-call audits, cost/latency monitoring, and failure handling.
- Select models across providers and partner with product and knowledge engineering to shape the AI roadmap.
Requirements
- AI or ML engineering experience, including shipping LLM systems for real users.
- Hands-on experience building and debugging multi-step, tool-calling workflows with LangGraph, LangChain, or equivalent.
- A repeatable approach to LLM evaluation, including datasets, regression testing, LLM-as-judge, or human review loops.
- Experience assembling context for LLMs and a clear point of view on retrieval strategies.
- Track record of owning systems from deployment through monitoring and incident response.
- Comfort working across model providers and explaining tradeoffs in quality, latency, cost, and risk.
- Comfort with Neo4j and Cypher, or a comparable graph store, with ability to ramp quickly on graph data modeling.
- Strong Python skills with production habits in FastAPI, asynchronous services, testing, observability, and maintainable interfaces.