Senior​/Staff Applied AI Engineer

About Tali

Clinicians are drowning in paperwork. We're giving them their time back.

Tali AI is one of the fastest-growing startups in Canada, on a mission to make healthcare more accessible with AI. We're building the clinical operating system: an ambient AI scribe, a billing agent, scheduling, clinical decision support, and more on one platform that allows clinicians to look at their patients instead of their keyboards. Thousands of clinicians across Canada and the US already use Tali, across dozens of specialties, integrated deeply with the fragmented landscape of North American health-record systems.

The traction is real: multiple commercial product lines, 15M+ patient visits documented on the platform, 70+ years worth of time saved for clinicians in under three years. We move fast in a market that rewards speed, in a domain where breaking clinician trust is not recoverable.

You own AI systems end to end. The models, the prompts, the retrieval, the data, the evals, and the debugging when a clinician says the output was wrong.

The ambient scribe listens to a visit and writes the clinical note. A recommender suggests the billing codes a clinician can claim for that visit. Medical search answers a clinical question from real sources. And a growing number of agents run inside the company and in the product. You work across all of it.

Evaluation carries the same weight here as any product feature. It is what makes the rest of it trustworthy.

• Build the evaluation pipelines that decide whether the AI is good enough to ship. Automated judges, regression suites, human review, and the datasets underneath them. On the audio path that means word and speaker error rates and the audio-quality measures that tell you a recording was worth trusting.
• Diagnose failure modes and fix them. The lever might be the prompt, the retrieval, the routing, the model, the audio capture, or a fine-tune.
• Build production agents. Tool use, orchestration, guardrails, and recovery when a step fails. You also build what sits under them: search, vector storage, and the harness the agents run in.
• Debug one visit end to end, then find every case like it. You trace a single interaction from audio to delivered note and work out what broke. Then you slice the warehouse to size the problem and prove the fix.
• Decide which model serves which request, and change that safely. Weighted routing, staged rollout, and attribution good enough that you know which change moved the number.
• Own these systems in production. You get the alert when quality slips, you find the cause, and you decide what ships to fix it.
• Turn a vague clinical complaint into a problem statement, a metric, and a plan the team can act on.
• Set the bar for how Tali does applied AI. Your evals become the evals everyone else runs.

• 5+ years in production ML, applied AI, or research engineering. You have owned something that ran for real users and stayed up.
• Deep evaluation experience. You have built graders, regression suites, or judge pipelines, and you know how to tell when a judge is fooling you.
• Agentic systems. Multiple models, tool calls, and retrieval, with the failure recovery that makes them safe to run.
• Strong systems engineering. Backend services, data pipelines, and enough observability that you can answer questions about production quickly.
• Data-centric instincts. You improve an AI system by improving its data and its feedback loops, and you can say when a prompt change is the smaller lever.
• Python, plus modern ML tooling. You write code others can run.
• Candour. You give hard feedback on a colleague's design, and you take it on your own without going quiet.
• You make the case for the harder right answer in engineering terms and in business terms, then you ship it and own the result.
• You raise the people around you. Your review makes the next engineer's system better.

This is a senior or staff role depending on your track record. The levelling conversation happens at the end of the interview process.

Bonus points

• Speech recognition or real-time audio.
• A regulated domain, such as healthcare or finance.
• Clinical experience of any kind.

Rigor…