Launch HN: Cenote (YC W25) – Back Office Automation for Medical Clinics
Hey HN, this is Kofi, Kristy and Ajani, co-founders of Cenote (https://www.joincenote.com/). We provide medical clinics with AI agents to speed up their referral intake.
Before a specialist physician can treat a patient, they must collect data about the patient, determine if the referral meets medical necessity, and see if insurance will cover the procedure. This involves analyzing referral documents, coordinating with primary care providers for missing information, and verifying insurance coverage—before they can even see the patient. It’s a manual back-and-forth process that is time-consuming, prone to errors, and slows down patient care.
Cenote mostly automates this workflow. (“Mostly”, because sometimes a human-in-the-loop is needed—more on that below). We use LLMs, OCR, and RPA to extract and validate referral data, check for medical necessity, and initiate insurance verification—all in minutes, not hours. This allows specialists to focus on care, reduce administrative burden, and ensure faster, more reliable insurance payments.
One of us (Kristy) dealt with this after an emergency medical event she had a couple years ago. The time it took her to find a clinic that could receive her medical record and insurance exacerbated her injury. It seemed crazy to have to wait that long for what turned out to be the dumbest of technical reasons. The three of us became friends at a book club, got talking about this, and decided to build software to deal with it.
Cenote automates the back office for medical clinics. When a referral lands in a specialist’s inbox, our software kicks in. We first parse the document through an OCR. After that, we use an LLM to detect the pieces of data that our customer has told us they’re looking for. If we detect the referral is missing data, we send a message back to the referring provider asking for more. Finally, we integrate with our customer’s EHR (Electronic Health Record) via RPA or API and place the document and extracted data in its appropriate location.
The OCR returns confidence intervals. If the LLM reasons over OCR that it is not confident about, we flag this in the UI to the end user and ask a human to review before moving forward.
We entered this task thinking we would have to work on a lot of fine-tuning / ML infra, but the tech needs turn out to be a lot more elementary than that. For example, we have spent a lot more time creating a history-page view of previously submitted files than we have spent training our own data. Many clinics still rely on faxed (!) referrals, and even well-funded practices use obsolete workflows.
While we provide a UI for clinics to upload documents and for human-in-the-loop intervention, our system can also function in a headless manner. By this, we mean that all core functionality—data extraction, EHR integration, and even back-and-forth communication with referring providers—does not explicitly require a UI for user interaction.
In terms of pricing, we charge an annual SaaS fee and a one-time implementation fee. We don’t have one-size-fits-all pricing on our website yet, but we’ll get there eventually.
If you have medical clinic experience, we’d love to hear your thoughts! And everyone’s feedback is welcome. Thanks for reading!
Emergency injuries should be handled in an ER, where patient care comes first. They won’t bother trying to get your insurance information until after you’ve been treated and stabilised.
As far as specialists go… when I go to a specialist, they key in my insurance card and have an approval within seconds. Of course with a serious injury I’d be at an ER not sitting around a specialist’s waiting room.
My biggest concern, though, is this will be used to replace back office staff and serious mistakes will get made, patients will be the ones stuck with figuring out insurance nightmares - there won’t be any back office staff left to help, and providers will be given heavier workloads with less assistance. And no, I don’t trust LLMs to make medical decisions.
> in an ER, where patient care comes first. They won’t bother trying to get your insurance information until after you’ve been treated and stabilised
speaking from 1st hand experience, you are wrong.
> this will be used to replace back office staff and serious mistakes will get made, patients will be the ones stuck with figuring out insurance nightmares
on this you're spot on!
In the screenshot on your site alongside the "No more typos" feature, you have a typo: "United Helathcare". Also an error in the 'extraction' of the phone number ("-55-").
Thx for pointing these out !
Do you not see the irony in that?
Good luck getting into brick and mortar clinics (I really mean it!). It is so incredibly hard to get established, small businesses to do anything regarding IT or tech. They are all incredibly overworked and the last thing they want to deal with is tech or learning how to use a new platform.
Yeah some of our first customers have been smaller/brick and mortar clinics! We find getting in-person time with these owners and personally offering to train staff goes a long way in ensuring trust and confidence in using our product. On that note we have found conferences and meet-ups super helpful.
Is part of your value proposition that your product will replace some of the staff?
What if it fails to mention a critically important piece of info? Would your company be liable or would I as a physician have be liable for its mistake?
Great question. Our software is designed to assist, not replace, the physician’s role in making clinical decisions. It accelerates the time between an inbound referral and patient care by extracting and organizing information, but the final review always remains with the physician.
To minimize risk, we implement safeguards to prevent hallucinations, and our system is built to flag potential missing or unclear information rather than override clinical judgment.
So I'd still need to do a full chart review? I'm not sure it would save me any significant amount of time.
When you say you integrate with EHRs using RPA or API, are you using FHIR for the API connection? Or what interop standards are you using?
We're doing both RPA and API integrations now - depending on what works best for any given EHR/clinic. FHIR connections are on the way.
[dead]
There are so many companies in the transcription -> EMR -> insurance automation space. What differentiates you?
In fairness, I think of EMR/EHRs as thin wrappers over insurance automation to begin with.
Most of them sell to the C-suite first (money, reports, and compliance) and due to that those areas often get the most focus. I believe it's why a large portion of EMR's suck balls.
So yea, teh central question in most systems isn't "Is this patient getting better" it's "Can I bill this visit?"
I’d actually love to hear more—can you expand on this point here?
Not sure if you are familiar with payor systems or not, but an EMR/EHR basically gathers all the information sent to a payor system and prepares it in the format the payor wants.
A good EMR/EHR does the same thing with referrals and authorisations, and bigger hospitals will have direct real time links to insurers to approve those referrals. Most of them happen instantly. The ones that aren’t require manual review from a medical practitioner who workers for the insurer.
I do not understand how a cobbled-together LLM based system will do this better than the existing EMRs.
For sure. At Cenote, we’re obsessed with ensuring our software delivers real value to clinics, rather than just adding another point solution—or worse, overwhelming them with multiple fragmented tools. As mentioned in another thread, many of these clinics aren’t the most tech-forward, and we've found that in-person discussions often reshape our bundle (e.g., prioritizing referral intake over insurance verification). This tailored approach ironically simplifies integration and maximizes ROI for our customers.
> rather than just adding another point solution—or worse, overwhelming them with multiple fragmented tools ...
> This tailored approach
It really is AI slop all the way down now isn't it?
Thanks for clarifying that you have little to no interest in enhancing the quality of outcomes or success of treatment or quality of life for patients! Very telling!
Not our intention at all. Let me know what gave you that impression? We generally see the incentives of the owners we work with as very aligned with providing the best quality of care to their patients, and see our software as accelerating delivery of that care
Good luck with HIPAA! Maybe, unsolicited advice and all, target patient recruitment CROs first? They seem to be “edgier” than big hospicorp…peace
> The OCR returns confidence intervals. If the LLM reasons over OCR that it is not confident about, we flag this in the UI to the end user and ask a human to review before moving forward.
This seems helpful, but what if the flagging system misses an error? Do you measure the accuracy of your various systems on your customer data? These are typically the more challenging aspects of integrating ML in healthcare.
This seems really interesting. I'm curious how this compares to fully automated EHRs similar to what Modernizing Medicine has built for Dermatology practices. Also there's a startup called DayDental which does RCM, for dental practices. Additionally, are you planning on integrating with large EHRs like Epic/Cerner, or is this for smaller EHRs like SimplePractice?
I appreciate the interest! We love the efforts at the federal level and by other tech companies to modernize healthcare. We see AI agents as the next step in this evolution—where nearly all back-office needs can be productized into AI, enabling Cenote to provide every clinic with a best-in-class back-office team.
To your latter question, we’ve spoken with many hospital networks using Epic that would benefit significantly from our software. However, integrating with larger EHRs is notoriously labor-intensive, so for now, we’re prioritizing more accessible clinics.
Large hospital systems with EPIC can often get near-instant insurance approvals. Interestingly, they’ve done this without using AI.
A few examples are Cleveland Clinic which has instant approvals for a wide variety of specialties with most of the insurances they’re panelled on. For another example, OhioHealth had both instant approvals and instant copayment/deductible estimates at the point of service back in… 2013 (at least with Medical Mutual).
Back office workers are skilled workers who often know how to do things like navigate an insurer who denies things they shouldn’t be denying. How is an automated system going to do that?
Same with my doctor/hospital system as well. They use Epic. I will request an appointment, and I know within a minute of clicking accept that insurance has accepted the appointment, and how much I will owe.
Only once did I not get approved (for a sleep study), so I called the doctor's office, and they got me approved within a couple more minutes after pushing something else, and I got a new estimate in my portal and via text letting me know it was covered.
If the insurance kicked back that appointment and some AI was responsible for getting it approved on the doctor end (AI is definitely used on the insurance end), who do I call?
I'm all for AI helping you out, possibly extracting useful information from paper forms, but we haven't used paper forms in a LONG time.
I'm not a doctor, but my wife is a Licensed Marriage and Family Therapist, and she's tried at my insistence to use some of the AI software for her practice, and it all falls flat to the point she will not try anymore and has sworn off AI completely. She doesn't use Siri, her browser blocks the google AI results, and most of her research is in her medical books anyway.
AI is the future, but today is the present.
Super cool ! What OCR are you using for confidence ? Tesseract? Curious how you flag for Humam in the loop.
As a full stack engineer + doctor, always good to see innovation and ideas in this space. Best of luck.
Thanks, we appreciate the support :)
Umm, how did you learn each profession and how many hours per week do you spend on each?
Can't speak for them, but its not super common so they're isn't going to be one answer that represents all doctors in this niche.
For me: Got CS degree in undergrad, worked as a SWE full time for for 2 years, did OSS and some consulting work in med school (0-20hrs/week). Now my work is primarily clinical as a resident ~65 hrs/week, with just 5-10hrs/week on programming.
Maybe I'm alone in this, especially on this site.
I'm beginning to become disillusioned with these things. We're replacing like 1000s of jobs with a system that will almost certainly do a worse job than before. And the money is split between hospital shareholders and VC.
I get that there's an efficiency (market) gain here. But these AI startups that target existing sector automations seem like they're most just attempting to drive wealth inequality in a period of already terrible westh inequality.
These are topics that we at Cenote also mull over. Right now how we see things are:
1. If we perform worse, we won't deliver any value to the owner and we'll soon be out of a business
2. It's our bet that AI agents can actually perform these monotonous, detailed tasks very well and that this will free up humans to take on higher value work.
3. That higher value work being: calling patients, educating them, helping facilitate patient care. This is ultimately the work the owners we talk to are excited for their teams to take on!
Depends on how you perform worse (if you do so). It takes a while for medical errors to occur (in human or automated systems ) and then there is a lag before consequences for the patient, and not all negative outcomes lead to complaints or lawsuits.
Is a world where nobody needs to work a utopia or dystopia?
At this point a world where no one needs to work would be dystopian. Are we going to rely on the benevolence of our increasingly for-profit government. On the benevolence of our oligarchs to allow us the labor the robots aren't capable of doing yet? I see the promise of post-scarcity, but I haven't seen anything close to the technology it would require. Just greed. Corner cutting and rent collection for profit. I'd rather not see our medical back offices enshittified.
The insurance companies have already enshittified the medical back office to levels beyond comprehension.
Yeah, but why "extract value" out of every note? Seems like a better application for open source and non profits.
You're not alone.
I work in healthcare and it's a sea of bullshit all the way down.
And most of the time they're being started by kids right out of school with ZERO clinical experience.
Put your seat belt on fellas. The future is coming when your stomach cancer will go unnoticed since AI is doing the work. A correction will happen and then these will get banned.
[dead]
[dead]
Cenote?
C Note?
$100?
Really?
A cenote is an inland tidal pool connected to the ocean. Mostly found along the Riviera Maya in Mexico.
Aha yes we were inspired by this latter definition, which explains the turquoise branding on our site.
And at least in my mind associated with Maya rituals.
Yes some of them are sacred sites for the Mayan peoples.
I'm thinking GP is saying that maybe human sacrifice rituals aren't the best look for a medical tech startup?
Perhaps not a good look, but oddly fitting given the current state of affairs.
Still have to change WebFlow's Entropy Site Template Title ;)