A doctor finishes talking with a patient. The microphone has done its party trick: there is a transcript, perhaps even a tidy note. Then comes the expensive question. Is a lab missing? Does the medication fit the formulary? Has a guideline changed? Is the patient quietly drifting toward a condition nobody has coded yet? RhythmX AI was built for that moment - the awkward beat after documentation, when information has to become a decision.
The Palo Alto company arrived in October 2023 with unusual starting fuel: $50 million from SAI Group and Deepthi Bathina, a former Humana chief product officer, as founder and CEO. The premise was not that a language model should play doctor. It was that a licensed clinician should get a better prepared second set of eyes, inside the workflow, while there is still time to act.
Its precision-care platform assembles a patient's longitudinal record with clinical guidelines, payer policies, social determinants, lifestyle information and financial context. Predictive models look for risk and disease trajectory. Generative tools explain recommendations in natural language. The physician gets a proposed next step, the rationale and the source guideline - then accepts, changes, investigates or dismisses it.
The four-beat clinical rhythm
The product is the missing verb
Most healthcare software is a noun factory: another note, dashboard, risk score or inbox. RhythmX wants a verb. Order. Refer. Recheck. Document. Route. The platform's public feature list includes risk profiling, a Patient360 data engine, disease-progression trajectories, personalized care recommendations and workflow intelligence. It can suggest labs, imaging, medication alternatives, follow-ups and the most appropriate clinician or channel.
That places RhythmX beside clinical decision-support and chart-intelligence companies such as Navina, Regard, Wellsheet, Bayesian Health and Atropos Health. Ambient AI businesses like Abridge, Ambience and Nabla are adjacent, but their original wedge was documentation. RhythmX's wedge is what happens next. The distinction is getting blurrier as everyone expands, which makes integration and proof more important than a clever category label.
“Clinicians get a clear next best action instead of just a summary or generic guidance.”Deepthi Bathina, founder and CEO
The newest expression is RhythmX Pulse, released in Microsoft Marketplace in July 2026. It sits behind Dragon Copilot. After Dragon captures the conversation, Pulse checks that encounter against the patient's record, live labs, care gaps, guidelines, payer requirements and formulary. Recommendations arrive in the workflow the clinician is already using. No ceremonial migration to Tab Number 19.
What $50 million bought
The launch capital did more than keep laptops open. RhythmX could draw on SAI Group's Eureka enterprise AI platform, healthcare assets covering longitudinal information on 300 million patients, and a wider organization that said it included 250 data scientists and 1,000 healthcare-focused staff. That is a serious head start for the unglamorous work of security, data normalization and clinical validation.
The recommendation and validation counts are company-reported activity measures. They show use and review, not by themselves a randomized clinical outcome or audited return on investment.
RhythmX sells to health systems, not to a worried person at 2 a.m. with a browser and a rash. The users are licensed clinicians in primary care and, increasingly, ambulatory, emergency, specialty and inpatient settings. The economic buyers are hospital executives trying to improve quality, clinician capacity, coding accuracy and financial performance. Pricing is private. The Microsoft listing offers a free trial, but enterprise terms stay where enterprise terms usually live: behind a meeting request.
For a clinician, the practical menu is broad but recognizable. Before a visit, the system can compress a long chart into the details likely to matter. During or immediately after the encounter, it can point out an overdue screening, a suspected condition, a treatment option, a more specific order or a medication alternative covered by the patient's plan. It can also flag documentation relevant to risk adjustment and suggest a referral route. None of this is meant to leap directly into the record without review. The point is to make the candidate action visible while doctor and patient can still discuss it.
For a health system, the pitch is a portfolio rather than a single magic trick: fewer missed care gaps, more consistent guideline use, less chart hunting, better coding support and more appropriate routing. Some of those benefits can pull against one another. An extra prompt may improve completeness while slowing the visit. A coding suggestion may help revenue while demanding careful audit controls. RhythmX therefore has to tune not only whether a recommendation is correct, but whether it arrives at the right moment and earns the scarce currency of clinician attention.
Sentara was the reality check
In February 2024, RhythmX announced a collaboration with Sentara Health, a large nonprofit system with 12 hospitals at the time. The work centered on personalized primary care: next-best actions from clinical and payer guidelines, earlier disease detection, EHR analysis and documentation support. Just as important, Sentara brought practicing clinicians and care-model redesign to the table.
This is the bit another founder can copy. Start with the painful decision, not the impressive model. Co-design with the people who will be blamed when it is wrong. Put the answer where work already happens. Show the evidence. Give the user a veto. Then measure the clinical and economic result separately from clicks, prompts and recommendations.
Editorial map of product emphasis based on published features - not a performance benchmark.
The first thing to fail is usually the handoff
The failure mode RhythmX has designed against is ordinary and expensive: a correct recommendation that appears outside the clinician's workflow can still fail. So can a model without current payer logic, a polished demo without enough longitudinal data, or a deployment that adds cognitive load while promising to remove it.
RhythmX's choices read like answers to those failure modes. It describes itself as EHR-agnostic but embeds in clinical workflows. Its privacy policy says the software does not autonomously diagnose, treat or take clinical action. Recommendations include guideline attribution. The licensed clinician remains in control. In high-stakes software, the rejection button is not an admission of weakness. It is part of the product.
Then came the distribution merger
In November 2025, RhythmX AI combined with Get Well, another SymphonyAI Group company, to form GW RhythmX. Get Well had spent roughly a quarter-century on patient engagement, navigation and care orchestration. The combined company said it served more than 150 health systems and 85 million patients, including 8 million U.S. veterans. Bathina became founder and CEO; Get Well founder Michael O'Neil became vice chair.
The terms were not disclosed, so this is not a tidy acquisition-price story. It is a strategy story. RhythmX had clinical intelligence aimed at the moment of decision. Get Well had patient touchpoints before, during and after care, plus long-standing health-system relationships. Put them together and the recommendation can travel: from pre-visit preparation to the exam room, care-team coordination and follow-up.
What changed was not the thesis that AI should personalize care. The company widened the delivery system around that thesis. In enterprise healthcare, the model can be excellent and still spend years waiting outside procurement. The merger gave RhythmX a larger installed path to buyers, workflows and patients. That may be the most reusable lesson in the whole story: when distribution is the bottleneck, do not pretend one more model release will fix it.
Steal this
Own the step after a crowded feature. Deliver inside the incumbent workflow. Pair every recommendation with evidence and a human veto. Borrow distribution through a partner whose customer trust already exists.
Do not copy this blindly
It breaks without reliable longitudinal data, deep EHR access, clinical validation, security work, change-management muscle and a buyer who can capture the financial value. A lightweight chatbot cannot impersonate that stack.
A promising machine still needs receipts
By the merger announcement, RhythmX said its platform had produced more than 300,000 personalized treatment recommendations. The 2026 Microsoft listing cites more than 170,000 validations by board-certified physicians. Those are meaningful signs that humans have put the output under a microscope. They are not the same as independently published evidence that the software improves outcomes, lowers total cost or pays for itself across different hospitals.
That distinction is not a scold; it is the next chapter. Clinical decision support lives in a harder neighborhood than note-taking. A recommendation can influence treatment, documentation, coding and trust. Buyers will want prospective evidence, subgroup performance, adoption rates, false-positive burden and clear economics. The company that wants to choose the next action inherits the obligation to prove the action was worth taking.
RhythmX AI's story is therefore less “AI replaces doctor” than “software finally reads the pile.” It has capital, a credible founder, a health-system collaborator, a human-in-control design and, through GW RhythmX, substantial distribution. The bet is that the valuable layer in clinical AI will sit between information and action. The remaining test is wonderfully unromantic: does it help a clinician notice the right thing, often enough, without creating a new thing to ignore?