CHAPTERS
Repeated YC rejections and the prototype that finally got them in
Tanay recounts applying to YC multiple times, getting rejected twice (including for being in high school), and finally earning acceptance after showing a working prototype. He frames early entrepreneurship as rapid learning driven by direct market feedback and customer proximity.
- •Rejected from YC twice; acceptance came with a working prototype
- •Early lesson: the market teaches faster than planning
- •Bias toward getting in front of customers and solving concrete problems
- •Early traction can feel like “we’re done,” but scaling is the real start
Commure + Athelas today: an AI operating system for providers
Tanay defines the company as a suite of products that automate clinical workflows, patient engagement, and the financial/revenue cycle layer for healthcare providers. He shares scale metrics (payments volume, visits documented, patient touchpoints) to illustrate platform breadth.
- •Commure Scribe for ambient documentation
- •Athelas Revenue Cycle as a “Stripe for healthcare” (claims, denials, appeals, BI)
- •Commure Engage for patient communication, reminders, procedure prep, bill explanations
- •Serves both clinicians and back-office revenue cycle teams
- •Operates at large scale across health systems and private practices
Why LLMs change healthcare operations: replacing outsourced admin work with software
The discussion connects recent improvements in LLM voice/text interaction to automating work historically done by large human teams (often outsourced/offshored). Tanay argues that shifting these tasks to software reclaims time for patient care.
- •LLMs recently became capable enough for real human-like interactions
- •Hospitals employ tens of thousands for repetitive administrative workflows
- •Legacy approach: offshoring revenue cycle work (R1 and others)
- •New opportunity: reimagine admin as pure software automation
- •Goal: redirect productivity back to patient care
Origin story: smartphone microscope + early computer vision for blood analysis
Tanay traces the initial spark to YC Hacks and papers on using cheap optics with a smartphone camera to mimic a microscope. He built an early malarial-cell classifier using then-state-of-the-art classical ML methods and turned it into a science fair project that evolved into a diagnostics concept.
- •Early computer vision era: segmentation + Random Forests
- •DIY smartphone microscope concept using cheap glass optics
- •Classifier for malarial vs non-malarial cells using internet datasets
- •Obsession with bringing lab-grade tasks closer to patients
- •Seed of Athelas as a med-device diagnostics company
Building a regulated medical device at 19: conviction, mentors, and early backing
Tanay explains why he and his cofounder believed they could build a real-world microfluidics/imaging device despite regulatory hurdles. Influential mentors and early supporters reinforced the belief that ML could tackle complex industries like healthcare.
- •Cofounder Deepika’s bioengineering/microfluidics background built confidence
- •Stanford ecosystem and mentors (e.g., Chris Manning, Richard Socher)
- •Mindset: no boundaries—go after complex, regulated industries
- •Early investor support (first check after YC) reinforced ambition
YC pace shock: first-principles clinical trials and an aggressive FDA plan
In YC, the team was pushed to challenge standard med-device timelines and strip processes down to essentials. They set a bold target: complete a clinical trial and prepare an FDA submission within the batch timeframe.
- •YC partner pressure to interrogate every “this takes months” assumption
- •“Zero-based budgeting” mindset applied to clinical trials
- •Goal: finish clinical trial and be ready for FDA submission by end of batch
- •Competitive drive: aiming to be best company in batch, not just best healthcare company
- •Learning to move fast inside a slow industry
Clinical trial in Juarez: hands-on operations to compress timelines
To bypass slow institutional pathways, they found a smaller, fast-moving partner hospital in Juarez, Mexico. They personally ran key parts of the trial, compared results against a standard lab system, and completed end-to-end execution in weeks.
- •Stanford’s quoted timeline/cost (years, $120k) prompted a rethink
- •Partnered with a small hospital willing to move quickly
- •Founders operated the trial themselves (sampling, running device, comparisons)
- •Benchmarked against Sysmex XE-5000 lab system
- •End-to-end trial execution completed in ~6 weeks
Post-YC med-device era: FDA clearance, manufacturing rigor, and first big contract
After YC, the company focused intensely on earning FDA clearance and hardening the device for real-world reliability testing. Commercial validation came early via a pharma customer using the device in a specific monitoring workflow, even before full approval.
- •18–24 months of singular focus on FDA clearance
- •Engineering for ruggedization (drop/kick tests, repeatable performance)
- •First major customer: pharma tied to Clozapine monitoring needs
- •Signed a ~$1M contract based on clinical results pre-approval
- •Early clinics adopted; FDA clearance marked the end of the first epoch
The “punch in the face” after clearance: scaling realities and discovering bigger pain points
After regulatory success and early revenue growth, the team realized the true challenge was scaling and expanding impact beyond a narrow patient segment. Deployments put them inside clinics where they observed many adjacent workflow and billing failures ripe for software solutions.
- •Revenue jumped from ~$0 to $3–4M quickly after clearance
- •Realization: clearance wasn’t the finish line—distribution and scale were
- •Device deployments created deep trust and access to clinic workflows
- •Observed manual steps (portals, faxes, pharmacist calls, claim paperwork)
- •Thesis shift: help 100% of a clinic’s patients, not only the initial niche
Pivot era (2020): moving from hardware-first to software platform
Faced with a choice between building a solid standalone med-device business or betting on a broader software TAM, the team chose expansion. COVID accelerated demand for remote monitoring and catalyzed the first software offerings that later broadened across the practice.
- •Fork in the road: best-in-class device company vs expanding TAM via software
- •COVID created immediate need for home monitoring and telehealth
- •Built telehealth portal + connected monitoring devices
- •Next wave: workflow automation for collections, claims, and payments
- •Started in a narrow cohort, then expanded organically across practices
Physician burnout and a new adoption dynamic: doctors become the buyers
Tanay argues prior healthcare IT digitization served compliance and billing, not clinician productivity. Post-COVID burnout plus clear time savings shifted adoption power toward physicians, including unexpected self-serve purchases that later expand into enterprise deals.
- •EHRs optimized for CFO/CIO/legal needs (get paid, don’t get sued)
- •Burnout creates urgency; clinicians push tools into institutions
- •Commure Scribe shows physician-led/self-serve adoption (credit card purchases)
- •HIPAA-compliant self-serve is possible; barriers often internal policy, not law
- •Bottom-up adoption can drive top-down integration and enterprise upsells
Ambient documentation at scale: what it is and the hard engineering underneath
Ambient scribing listens to patient-physician conversations and generates structured documentation for downstream billing and workflows. The chapter highlights why the product seems simple but becomes technically complex at large scale due to real-world connectivity and reliability constraints.
- •Pipeline: transcription → summarization → structured clinical documentation
- •Trained on prior documentation patterns and clinical source material
- •Massive growth in documented appointments (from ~100k to ~20–25M annually)
- •Hard problems: unreliable networks, offline mode, background uploads, robustness
- •Healthcare now adopting consumer-grade product expectations and reliability
Rebuilding the financial layer: revenue cycle as a cybersecurity-like adversarial system
Revenue cycle management is framed as an adversarial interface between providers and payers, where denial tactics and broken payer workflows must be detected and resolved continuously. Commure automates claim status monitoring, denial analysis, and appeals—sometimes via LLM-driven interactions.
- •“Stripe for healthcare”: payments, claims submission, denials, appeals, BI
- •Payers create many failure/denial modes; constant monitoring is required
- •LLMs can negotiate/appeal denials via voice/text workflows
- •Payer APIs often work ~80% of the time, requiring reconciliation and fallbacks
- •Core value proposition: physicians see revenue lift and operational relief
Platform strategy, bundling, and competing with Epic at national scale
Tanay explains why large healthcare outcomes require a platform approach: point solutions often plateau without distribution and expansion. He positions Commure’s speed, bundling, and physician-led momentum as differentiation, while noting Epic’s roadmap pace and economic impact on certain systems.
- •Healthcare tends to reward platform companies; point solutions hit growth ceilings
- •Bundling strategy likened to Microsoft’s historical playbook
- •Distribution is the hardest problem; revenue cycle can power broader platform GTM
- •Epic risk: native features can wipe out standalone “wrapper” point solutions
- •Some large profitable systems avoid Epic; Commure aims to be the system of engagement on top of EMRs
- •Forward-deployed engineering pods replicate early founder-in-clinic learning
The next decade of medicine: sensors, home-based care, AI copilots, and shifting incentives
The conversation closes with a forward-looking view: more care moves to the home via sensors and virtual workflows, while hospitals specialize in complex procedures. Tanay predicts AI copilots will surpass humans in many diagnostic tasks but remain constrained by regulatory/accountability structures, and he anticipates insurance evolving toward more catastrophic coverage.
- •Patient experience hasn’t radically changed yet; innovation is accelerating at edges
- •LLMs measurably reduce clinician time burden, potentially expanding access
- •AI-driven patient engagement can reduce no-shows and improve prep compliance
- •Future: pervasive sensors (wearables, home devices) + virtual care as default for low acuity
- •AI copilots for diagnosis/testing will grow; physicians remain accountable sign-off
- •Insurance may shift toward catastrophic coverage; routine care becomes simpler/out-of-pocket
- •Desired outcome: value accrues back to physicians/providers rather than payer layer
