Senior Backend Engineer, AI — Remote

We build a remote patient monitoring platform for cardiology and chronic care. Clinicians
use it to watch heart-failure patients between appointments: vitals stream in from
wearables, thresholds fire alerts, care pathways route follow-ups, and an AI assistant talks
to patients directly about their own health data. When our software is wrong, a clinician
misses something. That shapes how we work.

We're looking for a senior backend engineer who has shipped real LLM features — not demos —
and wants to own the AI surface of a clinical product.

What you'd actually work on. Our patient-facing assistant answers questions about a
patient's own vitals, sleep, activity and medication adherence. Making that safe is most of
the job: structured outputs that can't silently break, guards that stop the model stating a
figure it wasn't given, retrieval against a real medical article index rather than
model-invented links, crisis detection that routes to a human, and an eval suite that
catches a prompt change breaking a conversation it used to handle. You'd also work on the
clinical automation underneath — risk scoring, baseline calculation, threshold alerting —
and on the data layer that feeds it.

What we need. Strong Python and Django/DRF, at production depth. Solid PostgreSQL, and
comfort with a time-series store alongside it — our vitals live in MongoDB. Real experience
with Celery or equivalent, including why idempotency matters when tasks get redelivered. And
genuine LLM production experience: prompt design, structured output schemas, evaluation,
cost and latency, and a clear-eyed sense of how these systems fail. If you've only called an
API and been pleased with the result, this isn't the role.

We also care a lot about how fast you work with AI coding tools. Our team uses Claude Code
daily and it's a real multiplier — if you're fluent with it, say so and tell us how you use
it.

Compensation. 60,000–80,000 USD per year, depending on experience. Fully remote.

To apply, send your CV plus a few sentences on an AI feature you shipped to real users: what
it did, how it failed, and what you changed. That's the part we'll read first.