Senior Backend Engineer, LLM Systems

Note: We previously added this job under the title of Senior Software Engineer. If you applied to that posting, please re-apply to this one if you are still interested.

About CleverConsult

When a doctor in Canada refers a patient to a specialist, the patient's records go along with the request. These are called referral packages and they can run to dozens or hundreds of pages: faxed letters, old scope and pathology reports, years of lab results, notes from three other doctors. Somebody has to read all of it before the patient walks in, and usually, that somebody is the physician.

CleverConsult reads the package for them. Before the visit, our system pulls out what matters and writes a pre-appointment consult note. A pre-visit questionnaire collects the patient's own history and an AI scribe finishes the note during the appointment.

We started with gastroenterology because one of our founders is a practising gastroenterologist who lived this problem every clinic week. Our IBD Profile and IBD Dashboard turn those notes into structured clinical data that physicians use to track their patients with inflammatory bowel disease.

We're a small Toronto team and the product runs in real clinics. You'll report directly to the founder and own large parts of the backend.

Large language models are the core of the product

Most of what CleverConsult delivers is written by a language model. A physician signs the note so it has to be right. This role is mostly about building that model output. To be clear, we don't mean using AI assistants to write code. We mean designing, testing and running the systems where an LLM's output is the product.

That work includes:

• Breaking very long, messy documents into pieces a model can handle
• Writing and versioning prompts, and getting structured output the rest of the system can rely on
• Building evaluation sets and regression tests, so we know whether a prompt or model change made notes better or worse
• Catching invented facts, and linking each statement in a note back to the source page it came from
• Managing cost, latency and consistency across thousands of documents a day, and handling model upgrades without surprises

What else you'll work on

• The referral pipeline around the model: documents come in by fax, upload and EMR, and leave as a draft note in the physician's chart.
• Integrations with the systems clinics already use, including the Accuro EMR, patient questionnaires arriving by webhook, and fax numbers assigned to each physician.
• Structured clinical data: lab histories, medications and disease activity pulled out of free text, then fed into the IBD Dashboard and features like lab trend charts.
• The platform itself: performance, monitoring, deployment, and audit logging. We've completed our SOC 2 audit and handle health information under PHIPA, and you'll help keep both in good standing.
• You'll also review code, help set technical direction, and work directly with the physicians who use the product. When a note is wrong, you'll hear about it from the doctor who had to fix it.

Our stack

• Python and Django 5
• PostgreSQL and Redis
• TypeScript on Bun
• React
• AWS, Terraform and Docker
• LLM APIs

What we're looking for

• You must be based in Canada
• Several years building and running production backend systems, including at a senior level
• Hands-on experience shipping features where LLM output is what users receive: prompt design, structured output, evaluation and quality monitoring in production. Tell us about one you built, what went wrong with it, and how you measured whether it was working
• Strong Python. Django experience is a real plus
• Solid database design and SQL, with PostgreSQL in particular
• Experience building APIs and integrating with third-party systems that don't always behave
• Comfort with AWS, infrastructure as code, containers and CI/CD
• Clear writing and good judgment about when to ask, since we work remotely
• Nice to have: experience with health data, PHIPA or HIPAA, EMR integrations, or document processing and OCR