Show Notes
Your blood pressure spikes the moment the cuff goes on. You are sitting on crinkly paper in a cold room, and the number on the screen has almost nothing to do with the life you actually live.
Funding stage: Series A Plus. Hello Heart closed a $70M Series D in May 2022, has raised more than $138M total, and its own CEO calls it a “baby unicorn.” This is a mature, late-stage company, comfortably past the Series A threshold.
This is the white coat problem, and it is a tidy little metaphor for everything broken about reactive healthcare: we measure people at the exact wrong moment, in the exact wrong place, and then we wonder why outcomes lag.
Amir Dolev from Hello Heart spends his days inside that gap. Hello Heart is a preventive heart health platform built around a connected blood pressure monitor, a smart pill box, and a mobile app, and it is trusted by more than 150 Fortune 500 and government employers. The newest piece is Nia, which the company launched in October 2025 as the world's first AI heart health assistant. This episode is the rare founder conversation that hands you the blueprint instead of the brochure. If you are building vertical AI, health tech, or any agent where a wrong answer carries real consequences, this is the one to study.
The throughline is trust. Amir keeps returning to a simple idea: humans were never meant to be the data layer. The job of AI here is not to replace the doctor. It is to absorb the 90% of a visit that is administrative friction so the 2% that is actually human, the fear, the reassurance, the “how does this fit your life,” can finally breathe. He calls it the shift from reactive to preventive, and he is blunt that the only way to earn it is to build guardrails most teams skip.
Named Frameworks
The Two-Layer Guardrail Model
How Hello Heart keeps a medical AI safe enough to ship.
- →Development layer: a proprietary dataset of hundreds of likely questions paired with credible, accurate answers, used to continuously evaluate Nia before release.
- →Production layer: a separate monitoring agent that evaluates every Nia answer in real time and flags anything drifting outside the guardrails.
- →Human layer: licensed pharmacists who monitor responses and can intervene or take over a conversation when it needs to escalate.
The Professional, Not the Generalist
Hello Heart's deliberate stance on agent scope.
- →Nia is built like a specialist, not an everything-bot. Ask her who wins the NBA finals and she politely steers you back.
- →The bet: in regulated, high-stakes domains, a narrow agent that always ties back to the mission beats a charming generalist that drifts.
- →She does not hard-decline every off-topic question, she redirects, keeping the user engaged without abandoning the boundary.
The Symphony of Agents
Amir's vision for where this goes.
- →A single orchestrator, your phone, your car, eventually a household robot, that knows when to pull each specialist agent into the conversation.
- →Nia is one instrument. Inside her own heart-health domain, deeper sub-expertises get pulled in as the conversation goes wider or deeper: nutrition, psychological safety, and more.
- →Agent-to-agent communication: you arrive at your doctor already equipped with an agent that understands your conditions, so the human conversation can be about fears and feelings instead of data entry.
Signal Stacking
Why Hello Heart's data is hard to copy.
- →Start with blood pressure over time across many users, then layer medication data and symptom data to create one strong signal no single reading can produce.
- →An expanding signal set is on the roadmap: oxygen, weight (critical for heart failure, where sudden gain can flag a coming crisis), and cholesterol, aggregated through Apple Health and Google Health APIs.
Founder Experiment
Build your own version of the Two-Layer Guardrail Model for whatever AI feature you are shipping. First, write down the 20 hardest questions a user could throw at your product, including the dangerous ones, and write the ideal answer for each. That is your eval set. Then put a second, cheap model in production whose only job is to grade your main model's output against your boundaries before it reaches the user. You do not need a pharmacist on payroll to copy the principle: separate the thing that answers from the thing that checks the answer. Most teams ship the first and skip the second. Run it for one week and count how many catches the monitor makes.
Glossary
Preventive care
Treating risk and supporting health before a crisis, rather than reacting after a hospitalization.
White coat syndrome
Elevated blood pressure caused by the stress of being in a clinical setting, which can make in-office readings less representative than home readings.
Medication adherence
Whether a patient actually takes the prescribed medication, at the right dose, on schedule.
Guardrails
The rules and limits that keep an AI's outputs accurate, safe, and inside its intended scope.
Monitoring agent
A secondary AI that evaluates the primary AI's responses in real time to catch unsafe or off-scope outputs.
Multi-agent framework
An architecture where an orchestrator coordinates multiple specialized agents, each with its own knowledge, tools, and prompts.
Orchestrator
The top-level agent that decides which specialist agent to invoke for a given request.
Vertical AI
AI built deeply for one specific domain or industry rather than for general use.
Digital therapeutic
A software-driven, evidence-based intervention used to prevent or manage a medical condition.
Hypertension
High blood pressure, a leading risk factor for heart attack and stroke.
Q&A: What Founders Ask After This Episode
How does Hello Heart use AI in heart health?
Through Nia, an AI assistant launched in 2025, plus predictive risk models that have flagged abnormal blood pressure trends for years and prompted users to seek care.
What is Nia by Hello Heart?
Described as the world's first AI heart health assistant, a guardrailed agent that helps members understand their data, manage medications and side effects, and know when to escalate to a clinician.
How do you safely deploy a medical AI agent?
Hello Heart uses a development-time eval set of hundreds of question-and-answer pairs, a real-time monitoring agent in production, and licensed pharmacists who can intervene.
Should an AI agent be a specialist or a generalist?
Hello Heart's view is that high-stakes regulated domains favor a focused specialist that redirects off-topic questions back to its mission.
What is the future of the doctor-patient relationship with AI?
Routine data handling shifts to agents, freeing the human visit for emotional support and personalized decisions, with agent-to-agent communication preparing the conversation in advance.
Why don't people take their blood pressure medication?
Forgetfulness, fear of side effects, and lack of clear guidance when something changes, which Hello Heart addresses with a connected pill box, Nia, and pharmacist review.
Five Founder Questions This Episode Answers
- →How do I ship an AI product in a regulated industry without exposing myself to catastrophic liability?
- →Should my AI agent do everything, or should I deliberately keep it narrow and on-mission?
- →What kind of proprietary data actually creates a moat, and how do I stack signals to build one?
- →How do I design AI that earns user trust instead of just impressing them in a demo?
- →When AI absorbs the routine work, what is the genuinely human value my product should protect and amplify?
URLs Mentioned in the Episode
- Hello Hearthttps://www.helloheart.com
- Hello Heart launches Nia (press)https://www.helloheart.com/press/hello-heart-launches-the-worlds-first-ai-heart-health-assistant-nia
- AI Native Studenthttps://ainativestudent.com
- AI for Foundershttps://aiforfounders.co
- Inbox Alchemyhttps://inboxalchemy.co
