Data Scientist

Tellex | Healthcare Intelligence Fully remote, United States $145,000 to $170,000 base, depending on experienceAbout TellexEvery day, Americans write about their healthcare in public. We rate a hospital after a discharge, describe a wait in an ER, argue about a health plan on a forum, or review the place they work as a nurse. Almost none of it reaches the organizations being described, and none of it is captured by the patient surveys the industry currently relies on.Tellex collects that material at national scale, resolves it to named facilities, systems and clinicians, and turns it into intelligence that health plans and health systems can act on. Surveys ask a prompted question of a small, self-selected sample months after the fact. We read what people said on their own initiative, in their own words, about the care they actually received. That difference is the whole company.We are early and deliberately small. The dataset already runs to hundreds of millions of items across every major review platform, employer review site and health forum in the country, mapped against a registry of more than 100,000 US locations and around 2,000 health systems. The infrastructure exists. What we need now is someone who can make the analysis exceptional and make it land with a buyer.The roleYou will own the analytical layer of the product. That means the models that read the text, the evidence that they are right, and the findings that come out of them.Three things make this different from most data science jobs. The first is that our output is published and quoted, so it has to survive being wrong in public: methodology sections, stated limitations, and numbers you can defend to a chief quality officer who does not like the answer. The second is that you will be in front of customers. Health plan and health system executives will ask you what a score means and what they should do about it, and the quality of that conversation matters as much as the model behind it.The third is that implementation is no longer where the value sits. Models write a great deal of our code and we expect you to use them heavily. What we are paying for is the layer above that: choosing the right method, knowing the assumptions it rests on, spotting the answer that is fluent but wrong, and defending the decision afterwards. Anyone can now produce a classifier. Far fewer people can tell you whether it should be trusted.You will report to the founder. It is a small team, so the boundary between analysis, product and commercial work is thin by design.What you will actually doImprove the NLP models that classify sentiment, detect themes and aspects of care, and attribute feedback to the right organizationDesign the evaluation: annotation schemes, agreement measures, error analysis, and honest accounts of where a model breaksAnalyze model outputs across the national dataset to find patterns that are real rather than artifacts of who happened to postTurn those patterns into briefs, benchmarks and client-facing findings that are clear to a non-technical readerSit in customer conversations, explain the method, take the challenge, and feed what you learn back into the productSet the methodological standard for published work, including how we state limitations and handle self-selection biasDecide which parts of the pipeline are safe to hand to a model and which are not, and put verification around bothBring on and develop a junior data scientist, teaching them to interrogate an answer rather than accept one, and raise the standard of review across the teamWork with the founder on what gets built next and what we stop doingWhat we need from youEssential:Depth in the fundamentals: statistical inference, sampling and selection bias, evaluation design, and a real understanding of what the methods you reach for assume and where they failThe ability to choose an approach, articulate why it beats the alternatives, and hold that position under challenge from a customer, a reviewer or usRigor about verification. You should be able to look at a fluent, confident, wrong answer, from a model or a colleague or yourself, and identify what is wrong with itReal NLP experience: text classification, topic and aspect extraction, sentiment, and modern transformer-based approachesAt least two years working directly with commercial clients, presenting analysis and adapting it in response to what they sayExperience developing or mentoring junior analysts or data scientistsCompetent Python and SQL at meaningful scale. You need to be able to work in both and to read them critically, which matters more to us now than how much you produceFluency in cleaning, linking, validating and quality-assuring data of varied provenanceAWS, and comfort operating models in production rather than only in notebooksThe judgment to know when a finding is solid and when it needs another weekExcellent writing. You will produce material with our name on it that other people will quoteA degree in a quantitative, computational or scientific subject, or equivalent demonstrable abilityUseful, and worth mentioning if you have it:Familiarity with US healthcare: payers, delivery systems, quality measurement, CAHPS, or provider identifiers such as NPI and CCNExperience with record linkage or entity resolution at scalePublished research, or any experience of work that was peer reviewed or publicly scrutinizedPrior work at an early-stage company, with a clear view of what you liked and what you did notOn AI toolsWe use them constantly and we expect you to. There is no exercise here where you write code unaided, and no credit for doing by hand what a model does in seconds.The condition is that you understand and can defend everything you put your name to. If you cannot explain why an approach was taken over the alternatives, or walk through what a piece of code does and why it is correct, it is not finished, whoever or whatever produced it. Plausible output and correct output look identical until somebody checks. Here, the consequence of not checking is a published claim about a named hospital that will not survive contact with the people who run it.How we workRemote, asynchronous, and focused on delivery. We care what you ship and how well it holds up, not when you were at your desk. We review work seriously and give direct feedback. You will be expected to do the same in return.What you get$145,000 to $170,000 base salary, set by experience rather than by negotiationHealth, dental and vision coverage, and a 401(k)Paid time off with a floor of 20 days plus federal holidays, and an expectation that you take themAn annual budget for training, conferences and equipment, chosen by youNamed authorship on published research and briefs, and support to present themGenuine scope. You will define how this company measures patient experience, and that definition will outlast your tenureApplyingPlease apply directly via LinkedIn.