Senior Full Stack Engineer – AI Agents

Senior Full Stack Engineer (AI Agents)Full time | Remote, Canada melyn is a platform built for First Nations communities across Canada. Nations use it to run housing, infrastructure, governance and economic development from one system, and to secure the funding that pays for all of it. We add new Nations every week, in every province and territory. We are not a traditional SaaS tool. We are Service-as-a-Software: we deliver outcomes, not seats. Everything we build follows OCAP (Ownership, Control, Access, Possession), the First Nations principles governing how Indigenous data is collected, stored, used and protected. OCAP is not a checkbox here. It is an architectural constraint that decides where data lives and who can touch it. The roleThe platform already has a working AI chat assistant built on workflows and retrieval. Next stage: an agent for every module, each able to act on that Nation's data, coordinate with other agents on multi step tasks, and answer any question a staff member has about their community's records. These are real actions with real consequences for a government. OCAP means no community's data or traces can ever leave its tenant. Correctness, control and isolation matter more than raw capability. Guardrails are part of the design, not something bolted on later. You own this end to end. What you'll doEvolve the chat assistant from workflow plus retrieval into a multi agent system spanning every moduleDefine how each module's agent is bounded: tools it can call, data it can see, how we stop it acting outside its scopeBuild the orchestration layer for handoffs across modules, with state, retries and memory across turnsBuild the guardrails: input and output checks, prompt injection defence, PII handling, hard limits on what an agent does without human approvalTreat every tool call as a security boundary: least privilege, audit logs of what the agent read and did, safe handling of secretsImprove retrieval so answers are grounded in the Nation's own records, and the assistant says "I don't know" rather than inventing oneBuild the evaluation setup: measure false actions and unsupported answers against clear thresholds, under strict tenant isolationExtend the chat UI in Next.js: streaming, tool call status, approvals for risky actions, a clear picture of what the agent did and whyWork with the engineering lead to shape what gets built, review code, and help the team build with LLMsStack: Next.js on web, React Native on mobile, JavaScript and TypeScript throughout, Node.js and Python on the backend, REST APIs, PostgreSQL, AWS, models served through Amazon Bedrock. What we're looking for5+ years as a software engineer, including 1 to 2 years building with LLMs in productionStrong Python and TypeScript, with Node.js, PostgreSQL and React or Next.jsHands on with agent frameworks or your own orchestration (LangChain/LangGraph, Bedrock Agents, or similar)Shipped a chat based product: streaming, WebSockets or SSE, message state, making chat feel fastVector databases and embeddings (pgvector, Pinecone, Weaviate, or similar), plus RAG pipelines, chunking, retrieval and rerankingDepth in LLM orchestration and prompt engineering, and in extending agents with tools (structured outputs, tool schemas, permission models)You understand prompt injection, data leakage and tool misuse, and have built guardrails for them in productionObservability, logging and tracing for AI agentsCaching (Redis or similar) and queues or messaging (SQS, Kafka, or similar)Strong backend architecture instincts, and you can explain a tradeoff clearlyNice to have: multi tenant systems with strict data isolation, OWASP Top 10 for LLM applications, red teaming or adversarial testing of LLM systems, early stage startup experience. Why melynFully remote, flexible hours, health benefitsDirect collaboration with leadership, no layersReal ownership. You define how quality works here rather than inheriting someone else's processTechnology that makes a lasting difference for First Nations communities across CanadaAn early seat at a Service-as-a-Software company, in a category still being definedSmall team, direct communication, light process. How we hireApply through LinkedIn. There's an AI screening, then candidates will be invited to a brief conversation for culture fit and how you work. Then you'll meet with our engineering lead for a more in-depth conversation.