Truemed Is Hiring ML Engineers to Classify What Counts as Preventive Care
Truemed, founded by Justin Mares and Calley Means, is hiring a machine-learning engineer to help classify wellness products for tax-free health-account spending. Its Senior ML Engineer opening, described in a LinkedIn job post, puts the work at checkout:
“At the center of that is determining whether products qualify for pretax spending - a classification problem ...”
Truemed combines that classification work with licensed-provider review. A proposed purchase is assessed against medical-necessity criteria; a licensed provider reviews the customer’s information and may issue a Letter of Medical Necessity, or LMN, while an HSA or FSA administrator retains the final decision on reimbursement. The same market shift is pressuring traditional benefits administrators to automate eligibility and pushing wellness brands to put pre-tax payment at checkout; Truemed sits where those trends meet.
On its current site, Truemed said more than 2 million Americans shop with it, while the job post does not say what share of those purchases goes through automated classification. Zero G Talent’s live board lists the remote role as its newest addition within the past week. The title understates the work: the classifier sits between tax rules and a shopper’s access to pre-tax money.
The rest of the hiring board points to work beyond a standalone scoring tool. Truemed posted a Senior Software Engineer - Truemed Labs role, with a salary of $175,000 to $225,000 a year, and a Tech Lead / Manager - Merchant & API opening. Their titles do not establish that either role owns the classifier. They do show that Truemed is hiring across production software, merchant and API systems, and senior technical management.
A separate Product Manager, Merchant Payments & Integration opening makes merchant payment integration a product responsibility. The posting does not disclose whether the role owns eligibility logic.
Because Truemed’s eligibility rules point to Internal Revenue Code Section 213(d), a category label cannot decide a case. Two customers buying the same product can get different outcomes because their health profiles and intended uses differ. The classifier must account for that medical context, which is where the line between a general-wellness purchase and a qualified medical expense is drawn.
The Classifier Draws the Line Between Wellness and Medicine
A sensor is not a diagnosis
A fitness tracker can measure heart rate and sleep, yet it can remain a general-wellness purchase rather than a qualified medical expense for a health savings account, or HSA, or flexible spending account, or FSA. In guidance dated September 23, 2026, the Internal Revenue Service said that a wearable with medical and nonmedical uses does not become a medical expense merely because it tracks health data. For Truemed’s preventive-care classifier, a product’s capabilities show what the device can do, not why the buyer paid. The purchase qualifies when the buyer primarily uses it to assist with the diagnosis, cure, mitigation, treatment or prevention of a specific disease.
Eligibility therefore turns on the proposed purchase in context, not on a product category alone. The documented inputs include the item, the purchaser’s health context, and the reason for purchasing it. The IRS says a practitioner’s recommendation may link the information tracked to a disease, while symptoms, diagnoses, abnormal test results and risk factors may also bear on the determination. A catalog flag such as fitness or wearable cannot capture that record. If the model assigns one eligibility label before examining the facts, it would classify shoppers’ intentions rather than their expenses.
A Letter of Medical Necessity is the documentary hinge between those two tasks. Truemed places licensed practitioners in the approval path: customers provide health information, a practitioner reviews it and, when the purchase qualifies, issues an LMN. The documented workflow combines product details, customer health information, licensed-provider review and claims documentation. Public materials do not specify how those tasks are divided between software and clinicians.
General health has no shortcut
The IRS said:
“Medical expenses must be primarily to alleviate or prevent a physical or mental disability or illness.”
Section 213 of the Internal Revenue Code generally allows a deduction for qualifying medical care. IRS Publication 969 explains that qualified medical expenses can generally be paid or reimbursed through an HSA, FSA, Archer Medical Savings Account or health reimbursement arrangement. Using a tax-free account does not change the underlying definition. Expenses that merely benefit general health remain outside it.
Exercise shows the difference. The IRS said exercise does not qualify as a medical expense when a doctor recommends it but the shopper undertakes it only to improve general health. The recommendation supplies context; it does not change the purchase’s primary purpose.
A gym membership qualifies only when the buyer purchases it solely to affect the structure or function of the body, such as through physical therapy for an injury, or to treat a specific physician-diagnosed condition such as obesity, hypertension or heart disease. A weight-loss program must likewise treat a specific condition diagnosed by a physician. Calling a purchase preventive does not remove the general-wellness limitation.
Food and supplements require more specific tests. A special food cannot meet normal nutritional needs, must alleviate or treat an illness and must have a physician’s substantiation for the need; only the excess over a comparable product that meets normal nutritional requirements qualifies. A supplement must have a medical practitioner’s recommendation as treatment for a specific physician-diagnosed condition. The tax test itself is therefore item-specific rather than a broad flag for nutrition or wellness.
That boundary sits uneasily beside Truemed’s public framing. In April 2026, founder Justin Mares said, “Performance fitness and preventive health are the same conversation.” The tax framework draws a different line: a product’s wellness benefit does not establish medical necessity. Because eligibility depends on the customer’s health context, evidence of purpose has to travel with the purchase.
The letter is evidence, not a ruling
An LMN is a formal document from a licensed healthcare provider explaining why a patient needs a particular product, service or treatment for a diagnosed condition. The IRS’s medical-expense test requires a purpose beyond general wellness, and a generalized desire to improve health is not sufficient. A diagnosis-specific letter can support a claim, but it cannot guarantee payment or turn a personal purchase into a medical one.
Recordkeeping also differs by account type. HSA holders keep records, while for an FSA the plan administrator reviews claims before paying. Truemed’s eligibility guidance says its LMN supports payment or reimbursement, but the plan administrator makes the final determination.
The public materials do not say whether Truemed combines hard rules with trained classification. The documented outcome is case-by-case: a licensed clinician reviews the customer’s information, issues an LMN if the purchase qualifies, and notifies the customer if it does not. The materials do not reveal whether the machine-learning output is a single label or part of a multi-stage workflow. They also do not describe the software as diagnosing conditions, and a general-wellness goal or generic recommendation is not a substitute for a specific medical reason.
A practical test is a request to sleep better or lose weight without an underlying condition. It should not unlock a general-fitness purchase without a specific diagnosed condition, a medical purpose and supporting evidence. Without that chain, the system is classifying wellness marketing rather than tax eligibility.
Legacy Benefits Administrators Are Automating to Catch Up
Elevate said its AI claims automation platform can approve up to 78% of claims automatically, cutting processing time from 25 days to under two minutes. The product moves health-payments AI beyond answering questions and into claim decisions. Elevate sells to third-party administrators, health plans and benefits platforms that manage HSAs, FSAs and health reimbursement arrangements.
Navia and Truemed launched a deep technical integration to help Navia quickly process claims involving Truemed letters of medical necessity. PeakOne Administration’s partnership puts LMN access through Truemed’s practitioner network while PeakOne handles claims processing and communication.
HealthEquity is approaching the problem from the member side. Its agentic AI initiative is intended to help members optimize HSA contributions, maximize tax benefits and make informed decisions about healthcare spending and investment options, the company says. Separately, its HSAnswers educational tool answers questions about HSAs, FSAs and other consumer-directed benefits. HealthEquity said the tool does not generate responses outside the data it provides, so its answers stay within approved materials.
Those tools support decisions; they do not review medical necessity. The cited materials do not describe HSAnswers or the agentic initiative as issuing LMNs or deciding whether a fitness item, supplement or recovery product qualifies. Answering account questions does not tie a health condition to a product.
Lively is making a broader infrastructure bet. Its Axis platform, which Lively said uses agentic AI, connects accounts, plan rules, claims, payments, reimbursements, file management, reporting and employee support. That gives administrators one place to manage those records as claims move through the system. Lively’s public materials do not show that Axis makes the provider-backed medical-necessity judgment required by Truemed.
The cited materials do not document a product from the companies above that directly challenges Truemed’s provider-backed classification. “Catching up” describes their technical direction, not a declared response to the startup. Administrators control much of the machinery that processes and pays claims; Truemed controls an upstream screen. One product gap the research leaves open is a system that lets an administrator accept a merchant-classified wellness purchase and settle it without a manual handoff.
The old bottleneck was processing claims; the new one is deciding what a claim is allowed to mean.
On the evidence assembled here, the claims queue can become faster while the preventive-care gate remains upstream. The gate now appears where consumers choose how to pay.
Wellness Brands Turn Pre-Tax Dollars Into a Checkout Feature
Peloton put tax-advantaged spending into its checkout for selected fitness machines in October 2024, so qualified U.S. customers could save up to 40% by paying with HSA or FSA funds. The change turns reimbursement into a payment choice made before the purchase rather than a process that follows it.
After a shopper selects Pay with HSA/FSA funds, the checkout redirects to Truemed. The shopper takes a brief health survey and can complete payment through Truemed. A licensed medical provider then reviews the responses; if the shopper qualifies, Truemed provides an LMN. Reimbursement remains subject to the administrator’s rules, making this a targeted payment path rather than universal coverage for every wellness product.
Peloton’s current checkout guidance limits the offer to one-time purchases of the Bike, Bike+, Peloton Tread Flex, Peloton Tread Vision, Peloton Tread+ Vision and Row+. Rentals do not qualify. Customers can pay the full price with an HSA or FSA card, subject to approval, or split the payment between that card and a credit or debit card. For customers who do not qualify, Peloton says they can cancel before delivery.
Eight Sleep offers a related pre-tax payment option for its Pod 5. The company advertises an average saving of 30% when customers pay with pre-tax dollars. Eight Sleep also appears in Truemed’s merchant network, one of the brands through which the marketplace applies survey-and-LMN review to eligible purchases.
In December 2024, iFIT joined with Flex to offer HSA and FSA payment options for premium fitness products and services associated with NordicTrack and ProForm. Flex said HSA and FSA acceptance would bring in new revenue, increase cart sizes and improve customer retention. This was a payment-option partnership through Flex rather than Truemed.
Meta’s partnership, announced on April 17, 2026, brought the model into connected eyewear. The company’s announcement said consumers could buy prescription Ray-Ban and Oakley Meta AI glasses on Meta’s storefront with an HSA or FSA card, while Truemed supplied the payment technology. It was the first Meta product listed on the Truemed Marketplace. The payment option gives Meta customers another way to buy hardware that already requires a prescription; it extends Truemed beyond fitness equipment without treating every pair of glasses as eligible.
Three days later, NOBULL selected Truemed as its exclusive HSA and FSA payments partner for training footwear. Its announcement put Truemed’s network at more than 3,000 merchants. The release also said Truemed’s site had surpassed 1 million monthly visits and that more than $150 billion sat in HSA and FSA accounts, most of it unspent. The merchant count and monthly visits were Truemed-supplied reach figures; the $150 billion was a national account-balance figure, not an independent estimate of Truemed’s market.
For the shopper, the HSA or FSA card now appears at the end of a clinical workflow rather than the start of a reimbursement claim.
The ML Engineer Is the New Benefits Gatekeeper
The machine-learning hire marks a change in sequence: preventive spending reaches software before a benefits back office. It does not move clinical or plan decisions into the model; it moves the preparation of those cases upstream. Licensed practitioners still judge medical necessity, while plan administrators still decide what a claim receives.
The broader hiring plan spans production software, merchant and API systems, payment integration and product management, while the checkout documentation places customer information and LMN handling around payment. The public record does not establish that the ML model owns any of those steps. Its stated function is to classify a proposed purchase before clinical and plan decisions, not to supply either final decision.
The public postings do not disclose Truemed’s model architecture, training data, decision metric or whether the core system combines rules with trained classification. Nor is there cited evidence that the software diagnoses conditions. A defensible audit would need to connect each purchase’s eligibility inputs, the clinician’s LMN and the administrator’s final disposition.
If the classifier is used, a false rejection could block an eligible purchase, while a false positive would still face clinician review. Neither model output is a final plan decision.
In February 2026, PeakOne Administration partnered with Truemed so eligible participants could obtain an LMN through Truemed’s licensed-practitioner network while PeakOne handled claims processing and communication. In March, a NueSynergy partnership gave participants streamlined access to the Truemed marketplace, where licensed practitioners determine medical necessity for qualifying expenses. These documented roles keep medical-necessity review with clinicians and plan administration with benefits administrators.
In February 2026, Truemed said the Validation Institute had certified its marketplace process as the most clinically rigorous in the HSA/FSA market, with separate guarantees for clinical credibility and ERISA immunity. The announcement gave no accuracy rate for the classifier. Certification supports the existence of process controls; it does not grant blanket IRS approval for every wellness product.
The documented result is a split decision: a clinician says whether medical necessity exists, and an administrator says what the plan will pay. Software supports the classification and payment flow but does not hold either final approval. The ML hire sits upstream of that chain, making classifier design consequential to which cases enter review without transferring final medical or plan authority.
The job ad framed pre-tax eligibility as a classification problem. The harder test is whether each checkout decision can be audited from the eligibility review to the clinician’s letter and the administrator’s disposition. A machine can open the gate; the record has to show why.
Working in frontier tech? Zero G Talent tracks the openings: see every open Truemed role, browse frontier tech jobs, the companies hiring, and the people building the field.