What this organisation ships, and who runs it

Proven rule. Every body claim carries a tag. Anything inferred, single-sourced-and-shaky or contradicted without a winner goes to section 4. Estimates are labelled and show their arithmetic.

Kicker. MEDvidi's own requisitions — 60 live and 67 distinct non-clinical ones recoverable from the archive — describe an organisation with nine named units, not two tracks, and they name what each unit measures and what it considers failure in the company's own words. Read together with a de-facto shipping changelog reconstructed by diffing eleven captures of the patient-portal JavaScript bundle from February 2024 to the live 14 August 2026 build, three things fall out that no prior Finding has. One: the "AI Clinic Track" is officially the "AI Autonomous Clinic Track" — five product teams, four PMs, a stated goal of "a clinic that is fully autonomous, with the only manual touchpoint being the patient's appointment with their provider" — while the Core Track's own PM requisition names "integrate insurance billing carriers into the core funnel" as a core responsibility, and a separately-hired US Director of Revenue Cycle Management is building commercial insurance billing "beginning with Florida and California in 2026." The cash-pay company is becoming a billed company, and product owns it. Two: the compensation question is settled from public sources — MEDvidi published a $225,000–$250,000 base plus equity band for a US-employed VP of Operations/COO in March 2026, and the "Competitive compensation and equity aligned with impact" line in the brief Ilia received is copy-pasted verbatim from that COO requisition; the company's own August version of the VP of Product posting replaced it with Polish contractor benefits. Three: Smart IT has never run a public job board — zero career URLs across 2,059 archived smart-it.io pages — while MEDvidi's own engineering requisitions explicitly screen out agency people ("in a product company, not agency or outsourcing"). The 30+ engineers are MEDvidi's, not an agency account.

Source legend

---

1. What we found

1.1 The requisition corpus, and why it is the best public description of this org

MEDvidi runs Comeet as its ATS and exposes a public, unauthenticated API. A7 used it to count requisitions [PRIOR]; nobody read the bodies. Adding &details=true returns the complete text of every live requisition [ATS]. Separately, MEDvidi's WordPress careers plugin server-renders each requisition page, which means the Wayback Machine holds the full text of closed requisitions too [ATS-ARC]. Together these give the company's own written account of every non-clinical team it has built since March 2025.

Live, 2026-08-19 [ATS]:

The eight distinct non-clinical roles live today are: VP of Product (Contract, Management, Poland / Remote-EU / Remote), Engineering Manager (Management, 4 geos), Staff DevOps Engineer (Senior, 3 geos), AI Product Analyst — Voice AI Agents (Middle, 4 geos, filed under Engineering), Product Analyst (Clinical Guidance) (Middle, 3 geos, filed under Product Management), People Operations Specialist (Middle, 4 geos), and three evergreen talent pools — Product, Design, Engineering [ATS].

Archived, 2025-03-25 → 2026-08-15 [ATS-ARC]: 67 distinct non-clinical requisitions. Seven were renamed in place, which is itself evidence about how the org's language moved:

66.55A  "Senior Product Manager"                       12 May 2025
     -> "Senior Product Manager (AI)"                  12 May 2025
     -> "Senior AI Product Manager (AI Agents for Clinic)"  13 Mar 2026
        ... one requisition, open across ~12 months, renamed twice   [ATS-ARC]

DB.95E  "Senior AI Backend NodeJS Developer" -> "Senior Backend AI NodeJS Engineer"
29.56C  "Data Analyst (AI Track)"            -> "Product Data Analyst (AI Track)"
F8.161  "Tier 1 Support Agent EU" -> "Administrative Assistant EU (Medical Field)"
                                  -> "Medical Customer Support (Telehealth, Remote EU)"
48.86D  "VP Product"                         -> "VP of Product"                [ATS-ARC]

1.2 Nine named units, not two tracks — and what each says it measures

The posting says "two product tracks — Core Track and AI Clinic Track" [JD] [JD-SITE]. The requisitions name nine distinct organisational units, each with a stated mission. Every quotation below is MEDvidi's own copy.

The AI track's real name and shape. The Director of Product (AI) requisition (uid C8.E5F, first archived 2025-11-17, still live 2026-04-23) reads [ATS-ARC]:

"We are seeking a Director of Product who will join our AI Autonomous Clinic Track to lead 5 product teams. The track is responsible for the clinical workflow, support processes, and aspires to build a clinic that is fully autonomous, with the only manual touchpoint being the patient's appointment with their provider."

and, in Responsibilities: "Hire, coach, manage, and otherwise work with a team of 4 PMs in the track", "Lead the AI Autonomous Clinic Track together with the Director of AI", and "The Director of Product reports to the SVP of Product & Engineering" [ATS-ARC].

Three facts in one paragraph that the posting does not contain: the track's goal is stated as autonomy, not assistance; there is a Director of AI as a peer; and as of November 2025 the reporting line above the product directorate was an SVP of Product & Engineering (§1.8).

Core Track. The Senior Product Manager (Core Track) requisition (uid 3F.366, archived 2026-03-13) [ATS-ARC]:

"within Core Track teams focus on the key drivers of our business: how we acquire/match patients and how we keep them in treatment. PM in this role will own the patient's lifecycle from the first click to their long-term recovery."

Its Responsibilities carry two lines that reframe the business:

"Architect the automated logic that maximizes provider utilization. Your goal is to eliminate 'empty chair time' while ensuring patients are matched with the right provider in minutes, not days."

"Navigate the complex intersection of Growth vs. Regulation. You will integrate insurance billing carriers into the core funnel, understanding exactly how a change in booking logic cascades into clinical workloads and financial reconciliation."

That is the company, in March 2026, telling PM candidates that insurance billing is a Core Track funnel integration — not a finance project. It resolves A7's parked insurance thread [PRIOR] and pairs with §1.6.

The named teams, with first public evidence:

Two of these definitions matter more than the rest.

The Clinical Guidance team's own metric statement [ATS]:

"The Clinical Guidance (Experience Team) builds products that help healthcare providers work more efficiently and make better clinical decisions. Our mission is simple: Reduce administrative burden so clinicians can focus on patient care. … While AI powers many of our solutions, our success isn't measured by the sophistication of the models. It's measured by whether clinicians save time, experience less friction, and deliver better care."

The Automation team's own self-description [ATS-ARC]:

"the Automation team builds and operates a mix of AI and conventional product features that automate clinical and operational workflows — from AI-drafted medical documentation to automated resolution of patient issues, intake verification flows, provider tooling, and the backend pipelines behind them. … It's a full-stack product team, not an AI lab: we ship UI, backend workflows, integrations, and AI features side by side, and we measure them the same way."

Named cross-functional partners for that team: "Product, Engineering, Clinical, MedOps, and Legal" [ATS-ARC].

AI Ops & Growth, from its QA requisition [ATS-ARC]:

"a fast-moving, startup-style team focused on automating operational processes and maximizing revenue impact. This small team works at high speed, ships in days, and values experimentation. … automating MedOps workflows (slot management, supply/demand balancing, traffic redistribution) and, once those are in place, moving on to top-of-funnel growth experiments."

1.3 What the org says failure looks like

Three current requisitions carry an explicit "This role is not for you if" block [ATS]. This is the closest thing to a published anti-pattern list a candidate will find.

Engineering Manager — the most consequential of the three:

"This role is not for you if: You define EM success by delivery timelines and sprint velocity. You want to review code, make architectural decisions, or be the technical authority. You need a clear task list from leadership to operate effectively. Your background is primarily delivery management, project management, or agency work."

The December 2025 version of the same requisition (uid DD.06F) said the opposite: its responsibilities included "Technical Excellence: Raise the bar for engineering quality, best practices, and long-term technical health" and "Technical Support: Help teams overcome technical roadblocks" [ATS-ARC]. Between December 2025 and August 2026 MEDvidi rewrote the Engineering Manager role to remove technical authority from it explicitly. Somebody else holds it. Who, is A7's open question 4.1 [PRIOR], and this is dated, checkable evidence that the question is live rather than theoretical.

AI Product Analyst — Voice AI Agents [ATS]:

"This role is not for you if: You prefer working only with structured datasets and predefined metrics. You are uncomfortable analyzing raw conversations, transcripts, and AI-generated outputs. You expect clearly defined answers rather than investigating ambiguous problems. You are not interested in understanding how AI systems fail in production."

1.4 The AI-eval discipline, in the company's own words

verification/gaps.md §1 item 4 says the study has no answer to "what does AI evals in practice mean here" [PRIOR]. The company has published one. Three analyst requisitions describe it.

Voice AI Agents analyst — Responsibilities, verbatim [ATS]:

"Analyze production behavior of Voice AI Agents and conversational AI workflows. Investigate hallucinations, inconsistencies, silent failures, and quality degradation in AI-driven conversations. Work directly with call transcripts, logs, generated outputs, and operational datasets. Design evaluation frameworks for measuring AI quality, reliability, and business impact. Build monitoring systems and dashboards for AI performance visibility. Measure the impact of AI systems on operational efficiency, business outcomes, and patient experience. Detect quality shifts, anomalies, and emerging failure modes in production environments."

Requirements name "Understanding of LLM limitations, hallucinations, evaluation challenges, prompt sensitivity, and AI failure modes" [ATS].

Automation team analyst — the sharpest formulation [ATS-ARC]:

"You'll turn the production behavior of our systems into evidence the team can act on: which parts work, where they fail silently, and what to fix next. … Design evaluations for AI quality — both automated and human review — and close the loop into prompt and product changes. … Measure business and operational impact of automation initiatives and surface the risks they introduce." Environment: "ambiguous, fast-changing environments where ground truth often has to be constructed."

AI Track data analyst (Mar 2026) [ATS-ARC]: "Design and implement evaluation approaches to measure system quality and user experience", "Identify opportunities where automation improves outcomes, as well as areas where it introduces friction".

AI Ops & Growth QA nice-to-have [ATS-ARC]: "Experience testing AI/LLM systems, including hallucinations, response quality, prompt stability, and evaluation approaches (evals, guardrails)."

Read as a set: MEDvidi's stated eval practice is production-behaviour analysis owned by analysts, with human review, monitoring dashboards, and a loop back into prompts. What is conspicuously absent from every requisition: golden sets, labelled corpora, inter-rater agreement, a named regression gate, or any release-blocking criterion. The QA requisition mentions "CI/CD quality gates" only as a nice-to-have. So the discipline exists as observability, not as gating — which is precisely the gap a VP of Product could close, and it is defensible to say so in the room because it is read off their own job ads, not asserted.

1.5 The engineering stack and delivery model, published by the company

The Engineering Talent Pool requisition publishes the stack [ATS]:

"AWS (S3, Lambda, EC2, Aurora, EKS, Bedrock), Angular 21, Node.js, TypeScript, NestJS, PostgreSQL, Redis, Kubernetes, LangChain, LangGraph, Google Gemini, and AWS Bedrock, alongside AI-powered development tools such as Cursor and Claude Code. We follow a modern engineering approach with 2-week sprints, continuous delivery (including multiple deployments per day), and a high level of AI adoption across software development, testing, code reviews, simulations, and engineering workflows."

Areas it says it hires for: AI Backend, Frontend, DevOps & Platform, Data Engineering, AI/ML, Full-stack QA [ATS].

The Staff DevOps Engineer requisition adds the delivery machinery [ATS]: self-managed GitLab CI/CD with reusable templates and OIDC; Terraform and AWS CDK (TypeScript); Karpenter autoscaling and Kyverno policy on EKS; Prometheus / Grafana / OpenTelemetry / OpenSearch / CloudWatch; HashiCorp Vault as code; "blue-green deployments with automated, health-gated rollback"; "PostgreSQL zero-downtime schema migrations (expand/contract) and migration gating in CI"; "secrets hygiene… PHI-aware handling of logs and data". And a requirement no other company in this category publishes: "Hands-on agentic AI workflows (Claude Code or similar): delegating to autonomous agents and integrating their output into production."

Two implications. There is no infrastructure-maturity excuse available: a company already running blue-green with health-gated rollback and CI migration gating cannot argue that a release gate on AI quality is technically premature. And this is a first Staff-level DevOps hire ("you set the technical direction… you own the outcomes for reliability, cost, security, and developer experience across the company") — platform ownership is being created now, not inherited.

1.6 The US layer — three requisitions that reprice the whole job

MEDvidi hires two kinds of people: EU/Caucasus product-and-engineering contractors, and US employees for anything that touches the clinic, the regulator or the payer. The US requisitions are where the strategy is written down.

Director, Revenue Cycle Management (uid A7.A66, Remote-USA, Contract, archived 2026-05-10) [ATS-ARC]:

"The Director of Revenue Cycle Management will design, build, and operationalize MEDvidi's end-to-end commercial insurance revenue cycle, beginning with Florida and California in 2026 and expanding to additional states thereafter. This role owns every stage of the revenue cycle from eligibility verification through payment posting, denial management, and payer performance reporting."

It goes on to specify behavioural-health CPT codes to be billed — 90791, 90792 (psychiatric evaluation), 99213–99215 (E/M medication management), 90833/90836/90838 (psychotherapy add-ons) — CAQH profiles with a 90-day attestation cadence, credentialing platforms "such as Medallion or equivalent", MHPAEA parity appeals, place-of-service codes POS 02 / POS 10, and coordination with the CMO and General Counsel [ATS-ARC]. This is not exploratory. It is an implementation plan with named states and a year.

VP of Operations / Chief Operating Officer (uid 6C.468, Remote-USA, archived 2026-03-13) [ATS-ARC] — the single most useful public document about how MEDvidi pays executives:

"Compensation: Competitive base salary ($225,000–$250,000), meaningful equity participation, and a comprehensive benefits package."

and, in scope: "Act as the CEO's right-hand operator"; "Partner closely with the CMO and clinical leadership to ensure clear clinical authority boundaries… strong alignment between the Professional Corporation and MSO"; "Partner with Product and Engineering to operationalize AI-enabled workflows (clinical documentation, intake, support, scheduling, QA, back-office automation)"; "you will have access to an Executive Advisor for strategic guidance and mentorship" [ATS-ARC].

That last clause identifies the "advisors" A7 parked as unknown (§4.10) [PRIOR]: at least one named-in-writing Executive Advisor attached to the operating leadership.

Compliance Director (uid 56.168, Remote-USA, Full-time, archived 2025-12-07) [ATS-ARC]: US employment benefits (15 days PTO, 15 sick, healthcare from day 1, 401k after 6 months), "With a team of 200+ and rapidly expanding", a training remit including "ethical AI use in psychiatry tools", and "provide quarterly reports to the board and executives" — so a board exists and receives compliance reporting. Other US-side requisitions in the archive: Director of Clinical Practice Performance (Apr 2026), Medical Operations Manager — Regulatory Escalations (Mar 2026), RN QA Manager, Provider Relations Specialist, Medical Director [ATS-ARC].

1.7 Two compensation tiers, now sourced rather than estimated

A7 estimated the VP band from Polish market benchmarks with an admittedly unsourced FX rate and called it an estimate [PRIOR]. Two of its three inputs can now be replaced with company-published figures.

MEDvidi's own published US executive band (VP Ops / COO, Mar 2026):
    base $225,000-$250,000 + "meaningful equity participation" + benefits   [ATS-ARC]

MEDvidi's own EU leadership package (VP of Product, Aug 2026):
    "a competitive compensation package" (no figure)
    Contract, B2B, Poland / Remote-EU / Remote
    22 paid vacation days + 5 wellness days
    health insurance AFTER the probation period
    English lessons via Preply
    NO mention of equity                                                    [JD-SITE]

Benefit tiering across the whole EU requisition set, which encodes a ladder:  [ATS] [ATS-ARC]
    VP of Product, Engineering Manager .............. 22 vacation + 5 wellness
    Director of Product (AI) ........................ 21 vacation + 5 wellness
    Head of Product Design, Senior PM, Senior Designer, Staff DevOps, AI Ops QA
                                                     19 vacation + 4 wellness
    Product/AI analysts, People Ops ................. 19 vacation + 3 wellness

The equity line is not a recruiter invention — it is a copy-paste, and the company removed it. The "Why Join MEDvidi" block in the brief Ilia was sent [JD] appears verbatim on MEDvidi's own careers page in the 23 April 2026 capture of the VP of Product requisition [JD-APR26]:

"Opportunity to build and own the operational backbone of a growing health tech company · High-impact leadership role with direct influence on company trajectory · Close partnership with the CEO, executive leadership, and advisors · Ability to shape how AI and modern systems are applied to healthcare operations · Competitive compensation and equity aligned with impact"

The same five bullets, word for word, are the "Why Join MEDvidi" block of the VP of Operations / COO requisition [ATS-ARC] — a US, full-time, $225–250k + equity role. "Operational backbone" and "healthcare operations" are COO language, not product language. So the sequence is: the COO requisition was written first; its closing block was pasted into the VP of Product requisition in April 2026; and by 17 August 2026 the company had replaced it with the Polish contractor benefits block [JD-APR26] [JD-SITE] [REASON].

That changes the question to ask. Not "is the recruiter overstating?" but "your own careers page carried an equity line on this role in April and replaced it with a benefits list in August — what changed?"

1.8 The seat: three holders, and a search that has run for ten months

Combining this Finding's requisition archaeology with A8's medvidi.ai leadership captures [PRIOR] gives a complete, dated chain.

Julia Guzman        CPO                       listed to 2024-11-08; gone by 2025-03-19   [PRIOR/ARC]
Konstantin Valiotti SVP of Product & Eng.     on medvidi.ai Sept 2025;
                    "VP of Product, Engineering and AI" in MEDvidi's own 5 Feb 2026 release;
                    absent from the live leadership block                                 [PRIOR]
                    -- independently corroborated: the Director of Product (AI) requisition,
                       first archived 2025-11-17, states "The Director of Product reports to
                       the SVP of Product & Engineering"                        [ATS-ARC]
(vacant)            VP of Product             questionnaire authored 2025-10-16;
                    publicly listed at least from 2026-04-23 (London/Berlin/Remote);
                    relisted 2026-05-10 (Poland/Georgia/Remote-EU/Remote);
                    live 2026-08-19                                        [PRIOR] [ATS-ARC] [ATS]

Two corrections to A7 fall out. First, A7 concluded the public posting was "at most six weeks old" because Wayback showed no VP row on 2026-07-04 [PRIOR]. The requisition detail pages were archived under vp-product in London, Berlin and Remote on 2026-04-23, and under vp-of-product in Poland, Georgia, Remote-EU and Remote on 2026-05-10 [ATS-ARC]. The role has been publicly advertised for at least four months, in at least seven geographies, under two slugs. Second, A7's H3 framed the role as "a backfill that has been upgraded" [PRIOR]; the SVP title in the November 2025 requisition confirms A8's reading that the PM+Eng+Design+Data+PMM scope is a restoration of a seat that already existed, not an expansion.

Third, the search is genuinely long. Questionnaire authored 16 October 2025 [PRIOR]; predecessor still publicly quoted 5 February 2026 [PRIOR]; requisition public by 23 April 2026 [ATS-ARC]; still live 19 August 2026 [ATS]. Ten months from authoring, four months public.

1.9 Smart IT: no job board, and requisitions that screen agency people out

verification/gaps.md §2.3 asks whether the 30+ engineers are the VP's team or an agency account [PRIOR]. Four public facts answer it.

1. Smart IT has never run a public job board. The Wayback CDX index for smart-it.io holds 2,059 distinct archived URLs; zero contain "career", "job" or "vacan" [SIT-CDX]. The live site's navigation is About / Services / Industries / Cases / Blog / Contact — no careers entry [SIT]. A general web search surfaced no Smart IT vacancy board.
2. MEDvidi hires engineers directly, under its own name, on its own ATS, in Poland, Serbia, Spain, Portugal, Armenia, Georgia, Estonia, Cyprus, Italy and Turkey, on B2B contracts, with MEDvidi's own benefits schedule [ATS] [ATS-ARC].
3. MEDvidi's engineering requisitions explicitly exclude agency backgrounds. Engineering Manager requires "3+ years managing engineering teams in a product company (not agency or outsourcing)" and rejects candidates whose "background is primarily delivery management, project management, or agency work" [ATS]. The AI Ops & Growth QA requires "5+ years … in primarily product companies (not agency or outsourcing)" [ATS-ARC]. A company staffing itself through an agency does not write that.
4. Smart IT's own claimed MEDvidi team was eleven people. Its case card specifies "5 software engineers, 2 QA engineers, 1 BA, 1 PM, 1 UX/UI specialist, CTO (Executive-as-a-service model)" over 19–20 months [SIT] [SIT-CASE] [PRIOR]. That is the original build, not a 30-person standing account.

Defensible conclusion: the 30+ engineers are MEDvidi's own contractor organisation. Smart IT's current, documented roles are (a) the historical platform build, (b) the co-founder who is called MEDvidi's CTO everywhere except on MEDvidi's own site, and (c) executive search — both recruiters on the brief use @smart-it.io addresses [JD] [PRIOR]. What remains genuinely unknown is the contracting vehicle — which legal entity issues the B2B contracts (§4.2).

One further Smart IT artefact nobody in the Inquiry has connected: Smart IT publishes a case study for MDBerry, "a network of online medical marijuana clinics with licensed specialists in New York, New Jersey, and Rhode Island", built as a custom healthcare CRM with WebRTC video over Twilio STUN/TURN, "27 months of development (and counting)", "5 person fully dedicated team" [SIT-CASE]. MDBerry is one of the six brands in the dctr.app platform's clinicName registry that A1 read out of the patient bundle [PRIOR]. The testimonial on that page is attributed to a named CEO of MDBerry; that is a public professional fact and nothing further about that person is inferred or reported here. What matters professionally is the structure: the platform Ilia would inherit hosts at least one clinic brand that is a separate business with its own chief executive, built by the same vendor. Multi-tenancy is therefore not only a marketing convenience — it is at least partly a real second customer, which is the strongest public support the Inquiry has for its B2B question (§4.6).

1.10 Product velocity from outside: the method, stated honestly first

verification/gaps.md §2.8 proposed diffing the patient-portal bundle across Wayback captures to reconstruct a shipping changelog [PRIOR]. Here is exactly what was done, so a reader can audit it and a CEO cannot dismiss it.

Method. Pull the Wayback CDX index for join.medvidi.com* (194 objects: 79 JavaScript, 99 HTML) [BUN-CDX]. Download every archived JavaScript object with the id_ raw modifier, gzip- decompress where the archive stored the compressed body, and fetch the live build (whose last-modified header is 14 Aug 2026 12:01:45 GMT). Then compare contents, not filenames: Angular route-table literals (path:"…"), TypeScript enum members, and string constants [BUN].

Limits — three, and they are load-bearing.

1. Chunk coverage is uneven. Angular splits the app into lazily loaded chunks. Wayback captured the whole set on only some dates. Five capture dates are complete or monolithic and therefore comparable: 2024-02-24 (main + 17 chunks), 2024-04-15 (main + 9 chunks), 2025-12-01 (single 2.41 MB main), 2025-12-27 (single 3.62 MB main), and live 2026-08-14 (single 1.93 MB main) [BUN]. The 2024-06-23 capture is main-only, and the three routes that appear to vanish there (appointments, documents, prescriptions) were verified to live in chunk 617.f677fd0b4045ec38.js in February 2024 — so their absence in June is a capture artefact, not a deletion [BUN] [REASON]. The 2026-01-19 → 2026-06-24 captures are esbuild builds with partial chunk coverage: their route tables are complete (the router lives in main) but their string inventories are not.
2. Bundle size is not a velocity measure. The build system changed from webpack (main.<hash>.js) to Angular's modern esbuild builder (main-<HASH>.js) between 2025-12-27 and 2026-01-19 [BUN], and the talent-pool requisition says the stack is now Angular 21 [ATS]. Size before and after that migration is not comparable.
3. The patient bundle carries shared platform types. A1 already showed CRM channel enums sitting inside the patient app [PRIOR]. So the appearance of a domain concept in this bundle proves the shared model gained it; it does not prove the patient-facing UI uses it. Every claim below is worded accordingly.

Deployment cadence, as a floor. Nineteen distinct main build hashes across nineteen observation dates, 2022-12-10 → 2026-08-14 [BUN-CDX] [BUN]. Identical hashes appear only within the same day; every observation on a different day shows a different build. Shortest interval across a confirmed build change: 2026-01-19 → 2026-01-31, 12 days [REASON]. This is a lower bound only — it says the portal has been rebuilt at least roughly fortnightly, continuously, for 44 months. It is consistent with, and far weaker than, the company's own claim of "2-week sprints, continuous delivery, multiple deployments per day" [ATS]. Quote the company's claim; use the archive only to say it is not obviously false.

1.11 The de facto changelog

Routes (comparable snapshots only) [BUN]:

Dated to the tightest available window [BUN] [REASON]:

• Between 15 Apr 2024 and 1 Dec 2025 — the portal gains a membership route, a payment cards route, an account history route, a manage route, phone verification, and report-pharmacy-issue; and it loses standalone appointments, documents and prescriptions list views plus a start route. Read as one release stream: three separate list screens were consolidated into history + manage, the patient lost a dedicated prescriptions view, and recurring billing plus phone-based identity arrived.
• Between 27 Dec 2025 and 19 Jan 2026 — change-pharmacy/:id is removed. It has not returned. The live build carries the PostHog flag is_pharmacy_text_search_enabled [BUN], so pharmacy change appears to have moved from a dedicated screen to a flagged inline search inside another flow.
• Between 24 Jun 2026 and 14 Aug 2026 — auto-login and verification-code appear. Passwordless / link-based entry.

Domain vocabulary. Each token below was checked by direct search across every capture, not by inference [BUN]:

Four of these are worth stating out loud.

(a) The fill-status data exists, and has for two and a half years. verification/gaps.md §2.7 says two Findings nominate real-time pharmacy routing as the unclaimed moat and nobody checked whether the data exists [PRIOR]. It exists. The patient portal has carried a prescription status enum since 24 February 2024 whose members are UNKNOWN, ACTIVE, INACTIVE, DISCONTINUED, DELETED, COMPLETED, CANCEL_REQUESTED, CANCEL_PENDING, CANCELLED, CANCEL_DENIED, CHANGED, FULL_FILL, PARTIAL_FILL, NO_FILL, CANCELLED_MANUALLY, ALREADY_PICKED_UP, NO_RESPONSE [BUN]. That is the shape of an e-prescribing fill-status feed. MEDvidi is already receiving whether the pharmacy filled, partially filled, or did not fill. The unbuilt thing is not the data pipe; it is the product — routing, ranking and telling the patient before they drive there.

(b) The AI receptionist's voice channel is roughly eight months old at most. No voice or IVR vocabulary appears in any complete capture through 27 December 2025; a full inbound-telephony state machine is in the 14 August 2026 build [BUN]. That matches the AI Receptionist Team requisitions appearing in 2026 [ATS], and it means the "shipped AI" in the posting is, on the voice side, very new — which is an argument for eval investment, not against it.

(c) The in-portal chat SDK disappeared while the AI consent still discloses portal chat. Twilio Conversations client code sits in the December 2025 builds and is absent from the live build [BUN], while A1 records that the AI Usage Consent covers portal chat and that the route map has no chat route [PRIOR]. Either chat moved to a widget outside the bundle, or the disclosure runs ahead of the surface — a cheap, compliance-adjacent question.

(d) There is an outcomes model in the product. STABLE_HIGH, STABLE_LOW, IMPROVED_TO_LOW and IMPROVED_WITHIN_HIGH are severity-trajectory categories, present since at least December 2025 [BUN], alongside the Core Track requisition's "delivering treatment outcomes to our patients as our north star" [ATS-ARC]. No Finding in this Inquiry has an outcome number because the company does not publish one — not because it does not compute one.

1.12 The hiring manager's frame

Two long-form public texts by the CEO exist. Read them as a frame to align to, not as facts to verify. Everything below is his own wording.

What he says the company is. An operational-infrastructure company that happens to run a clinic. His origin story is not medical: in 2008 he joined a paper-based clinic where "it used to take about five days to locate and retrieve patient records", bought "a scanner and shredder", and concluded that "sometimes operational infrastructure is the foundation of good care" [CEO-INT].

What he says the constraint is. Not demand. "The biggest problem in healthcare is still providers' capacity. … within three months of joining us, most providers are 80% booked with follow-up patients" [CEO-INT]. "Up to 80% of psychiatric visits are routine follow-ups — often driven by regulatory requirements rather than clinical complexity", while "new patients wait 6 to 9 weeks" [CEO-INT].

What he says AI is for. The administrative layer, and nothing above it, for now. "AI's highest-ROI role in healthcare today is administrative. Using it too early in the clinical setting can create more problems than benefits" [CEO-ART]. "The right entry point is the administrative layer. Fix that first, showcase and prove safety, build trust, and then expand from there. That's the path MEDvidi is on" [CEO-INT]. His four named administrative use cases [CEO-ART]: chart generation, chart review, pre-visit workflow automation (identity verification, cross-checking state databases, history review, contraindications, drug-seeking screening), and routine prescription management.

Where he draws the line. "Crucially, the AI does not make independent decisions. That's the architecture we intentionally built." The AI is "a clinical verification layer, grounded in evidence-based guidelines and a proprietary dataset" [CEO-INT]. He puts a number on the training corpus: "MEDvidi AI is trained on 130,000+ real psychiatric visits" [CEO-INT] — a claim, undated and unaudited, but a specific one worth holding.

What he measures. Time returned to clinicians, coverage, consistency. "the AI Chart Generator transcribes visits in real time, updating documentation every 60 seconds, and cutting charting time by 10x. The AI Chart Reviewer monitors 100% of clinical encounters for SOP adherence, reducing chart review time by 80% while handling ID verification, drug-seeking detection, and guideline compliance" [CEO-INT]. Note his naming: AI Chart Generator and AI Chart Reviewer, not "AI Scribe" and "Chart Review AI". Using his names is free credibility.

What he is proud of, and how he pre-empts the review corpus. He volunteers the bad reviews before anyone raises them:

"That also means we sometimes say no. If a patient comes in expecting a specific medication because they read about it online, but it's not clinically appropriate, our providers won't prescribe it — and AI helps enforce that standard consistently. There is a tradeoff. Patients who don't receive the treatment they expect may leave negative reviews. But that's the cost of practicing responsible medicine." [CEO-INT]

He also states a governance body: the SOPs "are not created by business teams — they are developed and continuously reviewed by a committee of licensed medical professionals… fully auditable and can be reviewed by regulators at any time" [CEO-INT].

What he complains about. Three things, all product problems. (i) Data infrastructure: "Some standard analytics tools and data pipelines are not suitable due to HIPAA… You often need custom infrastructure from the ground up" [CEO-ART]. (ii) Clinician variance: "every clinician also brings habits, workflows, and protocols shaped by previous settings… AI has to account for that variation without drifting away from the standard of care. That is a much harder feat than building a model that performs well in a demo" [CEO-ART]. (iii) Bolting AI onto broken workflows: "The biggest mistake is trying to layer AI on top of broken workflows. …it should be rethinking where entirely new workflows can be built with AI" [CEO-INT].

Where he says this is going. "Utah has already opened the door to autonomous AI in prescription renewals through its regulatory sandbox" [CEO-ART]; "Over the next few years, we'll see fully automated follow-up care for stable patients. AI-managed visits with physicians in a supervisory role, confirming decisions. That's the standard we're trying to set" [CEO-INT].

Three things this implies for the conversation. First, he thinks in workflows and minutes, not funnels and conversion; a candidate who opens with CAC is speaking a second language. Second, he has already decided that AI must not decide — proposing autonomy will read as not having listened, whereas proposing evidence that autonomy is safe enough to earn reads as agreement. Third, he has pre-argued the negative-review problem, so raising the 1-star corpus as a quality complaint walks into a prepared answer; raising it as a measurement problem ("you say saying no is the cost of good medicine — do you know what fraction of declines happen after payment, and what that costs in refunds and CAC?") does not.

---

2. Capability / object table

---

3. Reconciliation notes

Reconciliation — how long the VP requisition has been public. A7 concluded "≤6 weeks public" from the absence of a VP row in the 2026-07-04 careers-page capture [PRIOR]. This Finding retrieves archived detail pages for uid 48.86D dated 2026-04-23 (London, Berlin, Remote, slug vp-product) and 2026-05-10 (Poland, Georgia-Tbilisi, Remote-EU, Remote, slug vp-of-product) [ATS-ARC]. Resolved in favour of the detail pages: the listing capture is a snapshot of one paginated index and can miss rows; a rendered detail page cannot exist for a role that was not published. The role has been public since at least 23 April 2026. A7's questionnaire-timestamp dating (authored 2025-10-16) stands and is unaffected.

Reconciliation — the equity discrepancy. A7 left this deliberately unresolved between "recruiter licence" and "a genuine negotiable the company chose not to publish" [PRIOR]. Neither is right. The exact five-bullet block including "Competitive compensation and equity aligned with impact" was on MEDvidi's own careers page for this requisition on 2026-04-23 [JD-APR26], and is word-for-word identical to the closing block of MEDvidi's VP of Operations / COO requisition, a US full-time role with a published $225k–$250k base and "meaningful equity participation" [ATS-ARC]. Resolved as: the block is boilerplate carried over from a US executive requisition, and MEDvidi replaced it with an EU-contractor benefits block between April and August 2026. The recruiter did not invent it; the company retired it. That is a better question and a fairer one.

Reconciliation — is there insurance billing? A7 recorded three converging signals and resolved only the direction, handing the question to A3 [PRIOR]. Two company-authored documents now close it. The Core Track PM requisition (Mar 2026) lists "integrate insurance billing carriers into the core funnel" as a responsibility [ATS-ARC]; the Director of Revenue Cycle Management requisition (May 2026) says the role will "design, build, and operationalize MEDvidi's end-to-end commercial insurance revenue cycle, beginning with Florida and California in 2026" [ATS-ARC]. Resolved as: cash-pay is the current retail offer; a commercial-payer line is an explicit 2026 build with named launch states, and the funnel work sits inside the Core product track. Not resolved: whether any claim has been submitted yet.

Reconciliation — what the two tracks are (gaps.md C10). A1 hypothesised Core = patient funnel/portal and AI Clinic = Doctor Cabinet + CRM; A5 derived AI Clinic = clinician guidance plus patient-facing agents; A7 found four named units fitting neither [PRIOR]. The company's own text settles it: Core Track = "how we acquire/match patients and how we keep them in treatment", containing at least a Patient Journey team and a Platform team [ATS-ARC]. AI Autonomous Clinic Track = "clinical workflow, support processes", five teams, four PMs, led jointly by a Director of Product and a Director of AI [ATS-ARC]. So A5 is right: patient-facing agents are in the AI track, because the AI track is organised by who does the work (a human or the system), not by whose screen it is. A1's clean surface-based split is wrong. A7's four units are four of the nine.

Reconciliation — three "removed" portal routes. An earlier pass of this diff appeared to show appointments, documents and prescriptions vanishing in June 2024. They were in fact declared in lazy chunk 617.f677fd0b4045ec38.js in Feb 2024, and Wayback captured no chunk on 2024-06-23 [BUN]. Resolved as a capture artefact. They are genuinely absent from the complete Dec 2025 and Aug 2026 builds, so the removal is real but dated only as after 15 Apr 2024, before 1 Dec 2025. Recorded because the method's credibility depends on catching this class of error.

Reconciliation — who the engineers work for. A7 parked this as the question that decides whether the VP can reshape the team [PRIOR]. Public evidence now favours one answer: zero career URLs in 2,059 archived Smart IT pages [SIT-CDX], direct MEDvidi-branded engineering hiring across ten countries on MEDvidi's own ATS [ATS], two requisitions that explicitly reject agency backgrounds [ATS] [ATS-ARC], and a Smart IT case study claiming an 11-person team [SIT]. Resolved as: the engineering organisation is MEDvidi's own. Still open: which legal entity issues the contracts (§4.2). That is a recruiter question, not a CEO question.

---

4. Open Questions / Parked

4.1 What is the fifth team in the AI Autonomous Clinic Track? Four are named across requisitions — AI Receptionist, Automation, Clinical Guidance (Experience), AI Ops & Growth [ATS] [ATS-ARC] — against a stated five teams and four PMs [ATS-ARC]. The arithmetic also does not close: five teams, four PMs. What would settle it: asking the CEO to name the teams in each track and who the PM is on each. It is a natural, non-confrontational opening question and the answer maps the whole org.

4.2 Which legal entity issues the B2B contracts, and does the VP hold hiring and firing rights? §1.9 establishes the engineers are MEDvidi-recruited; it does not establish who signs. What would settle it: the recruiters — a recruiter-stage question, not one for the CEO.

4.3 Is there still a Director of AI, and does that person report to the VP of Product? The Director of Product (AI) requisition names a Director of AI as a co-lead of the track [ATS-ARC]; no version of the VP of Product posting mentions one [JD] [JD-SITE] [JD-APR26]. Either the seat was folded into the VP of Engineering line, or it exists outside the advertised org. What would settle it: one question, and the answer changes who owns model and prompt decisions.

4.4 Does the Core Track have a director? The posting lists exactly 1 Director of Product [JD-SITE]; the only Director of Product requisition ever archived is explicitly the AI one [ATS-ARC]. If that requisition was filled, Core Track's four PMs report to nobody below the VP. What would settle it: asking who runs Core today. If the answer is "the CEO", that is the most important sentence in the interview.

4.5 Why was AI_CHART_GENERATION_ENABLED removed, and where did the Twilio chat go? Both are in the 27 Dec 2025 build and absent from the 14 Aug 2026 build [BUN]; benign explanations exist for each. What would settle it: "is AI Scribe on for 100% of providers now, or still gated?" — a normal product question with the same answer.

4.6 Is MDBerry an arm's-length customer of the platform or a related business? MDBerry is a tenant in the dctr.app clinicName registry [PRIOR], has its own Smart IT case study, its own clinic network in NY/NJ/RI, and its own named CEO [SIT-CASE]. Nothing further is inferred here. What would settle it: asking how many clinic brands the product org supports and whether any of them are external customers. That question is also the cleanest way into the B2B thesis A8 raised.

4.7 What is the actual PM-to-team mapping today? Eight PMs are advertised [JD-SITE]; nine units are named [ATS] [ATS-ARC]; the AI track claims four PMs over five teams [ATS-ARC]. The target design in §6 assumes a mapping that has not been verified. It is a design, labelled as one.

4.8 Are the six data analysts pooled or embedded? The Operations analyst requisition says "join our Analytics team" [ATS-ARC], which reads pooled; the Patient Journey and Clinical Guidance analyst requisitions describe an embedded partner to a specific PM [ATS] [ATS-ARC]. Both may be true at once, which would itself be the problem. What would settle it: asking who the analysts report to and who sets their priorities.

4.9 How many candidates have reached final stage in ten months? Requisition authored 2025-10-16 [PRIOR], public since at least 2026-04-23 [ATS-ARC]. What would settle it: asking. The answer is also an honesty test, because the April capture is public.

4.10 What replaced the SVP of Product & Engineering between February and August 2026? The predecessor was publicly quoted on 5 Feb 2026 [PRIOR] and the requisition was already public by 23 April 2026 [ATS-ARC]. Who has been running the function in the interim — the CEO, the VP of Engineering, or the co-founder — is not answerable publicly and is the thing most worth knowing.

4.11 Deployment frequency is a floor, not a measurement. The archive supports "at least fortnightly for 44 months" [BUN]; the company claims "multiple deployments per day" [ATS]. No public source adjudicates. Do not repeat the company's claim as verified.

---

5. What this does NOT cover

• The consumer funnel's behaviour and copy. The bundle diff here is used only for velocity and estate. What the screens say and how they persuade belongs to funnel-behavioural-design.md and demand-engine-seo-funnel.md.
• The clinician app and CRM changelogs. Wayback holds 22 objects for doc.medvidi.com and 6 for crm.dctr.app [BUN-CDX]; repeated CDX requests for those hosts were refused by the archive during this session after heavy use, so no diff was attempted. This is the obvious next hour of work and it is cheap.
• Unit economics, revenue, pricing. The insurance findings here are org and roadmap facts. What a payer would pay, and what it does to contribution margin, is A3's and the-number-set.md's.
• Regulation. The Utah sandbox the CEO cites [CEO-ART], the DEA instruments, and the 2026-12-31 expiry belong to regulation-risk-dea.md. Nothing here re-derives them.
• Clinical supply and licensure. The 36 clinical requisitions were counted, not read. clinical-supply-operations.md owns the panel.
• Any private-individual information. Every person named here is named in a role, by MEDvidi, by Smart IT, or by a publisher, in a professional context. No personal-life, family, residential or financial information was sought, found, or recorded, and no relationship between individuals is inferred from any similarity of names. Where a fact about a person is not published professionally, it is absent rather than guessed.
• Verification of the CEO's claims. §1.12 reports his frame. The "130,000+ visits", "10x", "80% reduction" and "$30M ARR" figures are his; ai-platform-and-b2b.md sets the rule for how far to trust that class of number and this Finding follows it.
• Compensation negotiation. §1.7 establishes two published tiers. It does not model what to ask for.

---

6. What this means for a VP of Product

The first thing to notice is that the company has already written the job description twice, and the two versions are different jobs. The posting says two tracks and 53 people [JD-SITE]. The requisitions say nine units, two of which — Patient Journey and Platform — sit under a Core Track with no director, and five of which sit under an "AI Autonomous Clinic Track" whose stated destination is a clinic where "the only manual touchpoint is the patient's appointment with their provider" [ATS-ARC]. A candidate who walks in describing the org as two tracks is describing the brochure. A candidate who names the nine units, says which of them overlap, and proposes five, is doing the job in the room.

The second thing is that the business is changing shape underneath the product org, and product owns the change. A cash-pay company hired a US director in May 2026 to build a commercial insurance revenue cycle "beginning with Florida and California in 2026" [ATS-ARC], and told Core-Track PM candidates two months earlier that they would "integrate insurance billing carriers into the core funnel" [ATS-ARC]. Every one of the study's competing agendas — Prime, pharmacy routing, the DEA continuity flow, pre-payment qualification — was argued on a cash-pay architecture. Eligibility checks, benefit verification, prior authorisation, denial handling and a second price for the same visit are not a finance workstream; they are the largest single change to the booking funnel since payment-before-intake, and nobody in the Inquiry costed it. Naming that, unprompted, is the highest-value sentence available.

The third thing is that the CEO has already told you what he will not buy. He believes the constraint is clinician capacity, the entry point is administrative, and "the AI does not make independent decisions… that's the architecture we intentionally built" [CEO-INT] [CEO-ART]. So the winning move is not to propose more autonomy; it is to propose the evidence system that would let him earn autonomy when Utah-style sandboxes open — which he himself flags as the direction [CEO-ART]. His own requisitions describe eval as production-behaviour analysis with human review and dashboards [ATS] [ATS-ARC], and mention release gates only as a QA nice-to-have. The gap between "we watch our AI" and "our AI cannot ship without passing" is the single clearest, cheapest, most defensible thing a new VP can close, and it is read off his job ads rather than asserted.

The fourth thing is that the org's shape is a discovery organisation wearing a delivery organisation's clothes.

People whose output is documents and decisions:
  1 Dir Product + 1 Dir UX + 1 Dir Data + 1 Lead PMM + 8 PMs + 6 analysts  = 18   [JD-SITE]
People whose output is shipped artifacts:
  30+ Engineers & QAs + 3 Designers                                        = 33   [JD-SITE]
Ratio                                                                     ~ 1 : 1.8   [REASON]
Engineers per PM                             30 / 8                        = 3.75     [REASON]
Analysts per data engineer                    7 / 1                        = 7.0      [REASON]
Designers per PM                              3 / 8                        = 0.38     [REASON]
Named units per PM                            9 / 8                        = 1.1      [REASON]

An organisation with roughly one PM per named unit and 3.75 engineers per PM does not ship fewer, larger things; it ships nine simultaneous small things and then hires an analyst to explain why none of them moved. And it keeps hiring analysts: seven distinct analyst seats were advertised in 2026 alone — AI Track, Operations, Patient Journey, Marketing, Automation, Voice AI, Clinical Guidance [ATS] [ATS-ARC].

The target org design

Stated as a design, not a measurement (§4.7). The principle is one team per system of record and per queue, not one team per feature, and the target is 6+ engineers per PM.

VP of Product
│
├─ CORE  (patient demand and continuity)              Dir Product: hire/appoint — currently unowned
│  ├─ Acquisition & Booking            PM 1 · Des 1 · Eng 6 · Analyst 1
│  │     owns: both consumer front doors (join.medvidi, start.ezcareclinic), ASRS test,
│  │           service/plan/state eligibility, slot selection, payment, phone verification
│  │     metric: session → paid → verified → attended; time-to-first-appointment;
│  │             decline-after-payment rate and its refund cost
│  └─ Continuity & Money                PM 2 · Des 1 · Eng 7 · Analyst 1
│        owns: end-of-visit rebooking, follow-up, membership, refills, pharmacy-issue
│              resolution, transfer-of-care, AND the insurance funnel integration
│        metric: RR1 (ADHD return rate), visits/patient/yr, % visits ending with next booked,
│                fill-success rate, first-pass claim acceptance once FL/CA go live
│
├─ CLINIC  (the work the clinician and the care team do)   Dir Product (AI) — in post
│  ├─ Clinician Workspace               PM 2 · Des 1 · Eng 8 · Analyst 2
│  │     merges: Clinical Guidance (Experience) + the documentation half of Automation
│  │     owns: charting/AI Chart Generator, AI Chart Reviewer, adherence, peer review, eRx,
│  │           schedule, provider finances
│  │     metric: minutes per note; note-omission rate on a labelled sample; flag-closure rate
│  │             and time-to-closure; provider utilisation ("empty chair time")
│  └─ Patient Operations Agents         PM 2 · Des 0.5 · Eng 8 · Analyst 2
│        merges: AI Receptionist + the patient-issue half of Automation + AI Ops & Growth's
│                MedOps automation
│        owns: SMS/voice/chat/email agents, CRM task queue, intake + PDMP verification gate,
│              refunds and disputes, slot/supply balancing
│        metric: containment rate by intent; escalation precision and recall; verification
│                cycle time; contact-to-pay conversion; cost per resolved contact
│
├─ PLATFORM & TRUST                     PM 1 · Des 0.5 · Eng 7
│     owns: Website x Service x State x Plan config, admin console, brand tenancy,
│           identity/auth, the video app, credentialing docs, PDMP credentials,
│           audit trail and the AI evaluation harness
│     metric: days to open a state; days to launch a brand; config-defect escape rate;
│             video call quality distribution; % AI releases blocked by a failing eval
│
├─ DATA PLATFORM        Dir Data · 2 data engineers (one is a hire) · analysts embedded above
│     owns: one semantic layer, one definition of patient / visit / active / churned, across
│           all brands; the labelled corpora and golden sets the eval harness runs on
│
└─ PMM                  Lead PMM
      owns: the WordPress demand engine as a product interface, pricing and packaging comms,
            provider-facing marketing

Totals: 8 PMs (exactly the advertised eight), 6 analysts embedded rather than pooled, 36 engineer-slots against an advertised "30+" — so the design fits only if the "+" is at least six, and if it is not, Platform & Trust drops to five and Acquisition to five [REASON]. Design is the one line that does not fit at any reading: the allocation needs four designers and the org has three [JD-SITE]. One net data-engineer hire.

What is over-resourced. Analysis. One analyst per PM and seven analysts per data engineer [REASON] is the signature of an organisation that answers questions by hand, and it keeps hiring in that direction [ATS]. The fix is not fewer analysts; it is embedding them against a shared semantic layer so their answers agree, and moving the marginal hire from analyst to data engineer. Management is also over-built relative to craft: three directors over eight PMs, three designers and seven analysts, with no director over Core (§4.4).

What is unowned. Five things, and every one of them is defensible to raise: 1. The marketing site. 476 WordPress URLs are the only acquisition channel, five state pages name the wrong state, and pages governing money are three years stale [PRIOR] — with no product team and no requisition anywhere in the archive [ATS-ARC]. 2. The second consumer brand. EZCare is live, on the same P.C., with its own GTM container and a different state list including New Jersey and Oklahoma [PRIOR]. Not one requisition names it [ATS-ARC]. 3. The video app. dctr.app/v/ carries Twilio call-quality thresholds and "provider-side technical failure" is a named full-refund condition [PRIOR], yet no team, requisition or metric mentions it [ATS] [ATS-ARC]. 4. Pharmacy fulfilment. The fill-status enum has existed since February 2024 [BUN] and there is a report-pharmacy-issue route, but no team is named for it and the dedicated change-pharmacy screen was deleted in January 2026 [BUN]. 5. AI evaluation as a gate. Owned by analysts as observation; owned by nobody as a release criterion (§1.4).

What to consolidate. Merge Automation into two halves — documentation into Clinician Workspace, patient-issue resolution into Patient Operations Agents — because its own requisition describes both [ATS-ARC] and a team that owns both owns neither. Merge AI Ops & Growth's MedOps automation into Patient Operations Agents and hand its "top-of-funnel growth experiments" phase to Acquisition & Booking; a team defined as "operational automation, then growth" [ATS-ARC] is two teams with one backlog. Fold Clinical Guidance (Experience) into Clinician Workspace.

Named product units today                     6 (+3 functions)          [ATS] [ATS-ARC]
Engineers per product unit, today   30 / 6  = 5.0                       [REASON]
Product units after consolidation             5
Engineers per unit after            36 / 5  = 7.2                       [REASON]
PMs per unit after                   8 / 5  = 1.6  (two units run a PM pair)

No engineer is hired to get there; the gain is that five teams each own a queue end-to-end instead of six teams sharing three of them.

What not to touch, and say so out loud. Clinical protocol content — the CEO says the SOPs are owned by "a committee of licensed medical professionals" and are "fully auditable" [CEO-INT], and that boundary is the company's best regulatory asset. The clinician recruiting machine — 36 of 60 requisitions [ATS]. Brand consolidation — EZCare is live revenue and merging brands is a six-quarter programme. And the engineering stack, which already has more delivery discipline than the product org has decision discipline [ATS].

The question to open with, and why it works. "Your requisitions name nine teams — Patient Journey, Platform, AI Receptionist, Automation, Clinical Guidance, AI Ops & Growth, Analytics, Design, Business Ops. The posting says two tracks. Which of those nine still exist, who's the PM on each, and which two would you merge if you had to merge two?" It is answerable, it is flattering in the right way — it shows the work without claiming to know the answer — and whichever two he names tells you what he already thinks is broken. Follow it with the one nobody else will ask: "Your Core PM requisition says that role integrates insurance billing carriers into the core funnel, and you hired a Director of RCM in May to launch commercial payers in Florida and California this year. Which product team owns eligibility and prior auth in the booking flow — and does the $195 price survive it?"