Location: Toronto, Ontario
You will sit inside a small delivery pod rebuilding an investigative analysis platform used daily by law enforcement and forensic teams. The product ingests enormous volumes of device data and has to turn it into defensible conclusions, so the agentic layer you build is not a demo: it plans, calls tools, checks its own work, and leaves an auditable trail behind every decision.
This is hands-on work with a hard deadline and a real user on the other end. You will design the harness around the model rather than the model itself: prompts, tool interfaces, context and memory, permission boundaries, runtime controls, and the evaluation suite that proves the thing behaves. You will hunt the failure modes that actually matter here, hallucination, looping, prompt injection, unsafe tool calls, and build the guardrails that stop them reaching a case file.
Expect to be in the room with the client, not behind a ticket queue. You will turn a half-formed requirement into an architecture, ship a prototype in days, then harden it into something that survives an audit. The team is senior, opinionated, and small enough that your judgment shows up in the product.
Chain of custody is sacred in this domain. Data that lands in a report two years from now must be provably the same data that came off the device. If you find that constraint interesting rather than annoying, you will do well here.
Toronto based and in the office. Candidates elsewhere on the East Coast can work, but a daily 8am to 10am Eastern overlap with the client team is required, so the West Coast is not viable.
Must Have Skills:
Production experience shipping AI-powered software, not prototypes. Strong TypeScript or JavaScript, with Python a close second. Real agent-loop work: planning, tool selection, execution, observation, recovery. Experience designing evaluations that measure task completion, accuracy, reliability, safety and cost. Comfort with graph-based orchestration and state machines for multi-step flows. Solid fundamentals in APIs, distributed systems, databases and cloud. Instrumentation instincts: you trace decisions and diagnose failures rather than guess. Client-facing communication, since you will be in front of stakeholders regularly. Able to work onsite in Toronto and travel occasionally. Must clear a background check.
Nice to Have Skills:
Agent frameworks, workflow engines or coding agents. Tool interfaces, execution sandboxes and human-in-the-loop controls. RAG, embeddings, vector search, fine-tuning. Adversarial testing, golden datasets, simulators. AI observability including trace and cost analysis. Multi-model or specialist-agent architectures. Prompt injection and data leakage mitigation. AWS, Azure or GCP, containers, Kubernetes, CI/CD. Open-source contributions to agent infrastructure. Regulated or high-assurance environments. Consulting or professional services background. Existing security clearance, or TribalScale experience within the last three months, either of which removes weeks from onboarding.
