How it works / How the technology works
How does the Rarus Health technology work?
This page is for technical readers, such as technology leaders at partner companies and investors who run technical due diligence. It explains how the system turns messages, documents and check-ins into care signals, and what Rarus Health commits to, for each of the five parts that the How it works page describes.
How is the Rarus Health system built?
Rarus Health built its own system around large language models. The models supply language. The Rarus Health software supplies the facts, the rules and the record. Rarus Health can replace a model without changing the rest of the system.
The person
On WhatsApp
Text messagesVoice notesPhotos and PDFsButtons in questionnairesThe AI agent, for conversation
Works through Rarus Health tools
Patient recordQuestionnairesKnowledge libraryDocument filingHealth signals and emergenciesConsentThe back office
Work that runs in the background
Document and speech readingPrioritizing health signalsPatient narrativeThe Care Managers
Web console
Drafts to review and sendLive takeover of any conversationNarrative and record for the care teamUnderneath all four: one shared record. It holds every conversation, a record of every action, the patient record and the verified library.
Part 1. The everyday conversation
How does the Rarus Health AI agent hold a conversation?
The Rarus Health AI agent is the conversation agent that Rarus Health built for WhatsApp. It answers from verified content, asks structured follow-up questions and flags signals that need a person.
The AI agent sees the recent conversation, the open tasks and what the record holds about the person. The record supplies facts such as the person's name and the emergency numbers for the country. So the AI agent never has to guess them.
A Care Manager can take over any conversation at any moment. The AI agent steps back while the Care Manager writes, and continues when the Care Manager hands the conversation back.
The AI agent also asks structured questionnaires as a conversation, one question at a time. Programs can run validated questionnaires this way. Program authors create new questionnaires as content, without new software.
Part 2. Answers from a verified library
How does the knowledge library answer questions?
The AI agent answers questions about diseases, rights and treatment funding from a library that Rarus Health writes and a person verifies. The AI agent does not answer these questions from a model's general knowledge.
Rarus Health does not split documents into fragments for search. Fragment search often separates a fact from the conditions that make the fact true. In health information, those conditions matter. So every answer draws on whole articles.
The system records every question that the library could not answer. The content team writes new articles from that list. So the library grows from the questions that families really ask.
Part 3. Documents and voice notes
How does the system read documents and voice notes?
Rarus Health turns what families send, including photos, PDFs and voice notes, into text that the record can use.
A family sends a photo of a lab report or a PDF of a discharge letter. The AI agent confirms receipt. The system then reads the document, files it in the person's record and sorts it into a category, such as a lab result or a prescription.
A family sends a voice note. The system turns the note into text, in the language the person speaks. The AI agent repeats what the system heard, so the person can correct a mistake.
Part 4. The patient narrative
How does the system write a patient narrative?
The system writes each patient narrative from what the record holds, and keeps the narrative current as the record changes.
The narrative draws on the person's profile, questionnaire answers, the caregiver's situation, filed documents and recent health signals.
Each new version continues the previous one. So the narrative keeps the history of the journey instead of starting over.
Part 5. Health signals
How does the system handle a health signal?
A health signal is a care signal about the person's health: something the person mentions that the care team should know about. Examples are a new symptom, a missed treatment or a visit to the emergency room. The system makes sure that the Care Manager sees every health signal, with a priority and a draft reply.
The AI agent flags a health signal whenever a person mentions one. The person does not need special words or a form. The AI agent can flag a signal even in the middle of a questionnaire.
The system then summarizes the signal in the person's own terms, sets a priority for the Care Manager and drafts a reply. Several messages about the same event reach the Care Manager as one signal.
The system works from coordination guidance for each disease. The guidance says which signals the care team wants to hear about at once.
A possible emergency follows a separate, faster path, which points the person to the local emergency services.
How does Rarus Health keep the AI safe and accountable?
Rarus Health treats the AI as a component that must earn trust with evidence. The engineering team tests the AI, limits what it can do and records everything it does.
The team tests the agent before each release.
Automated suites run scripted people of different ages, moods and needs against the agent. The suites also replay real incidents, so a fixed problem stays fixed.
People approve what matters.
A Care Manager approves every drafted reply to a health signal. A person verifies every library article before the AI agent can use it.
The software enforces consent.
The system checks consent before it collects information. When a person writes STOP, the system stops writing to that number. The model cannot override either rule.
Personal information stays minimal.
The library search never receives personal information. Partner companies receive aggregated, anonymized insight only. A person can ask Rarus Health to erase their information.
Every action is recorded.
The audit trail records every agent action. The system also records each time a staff member opens a patient's information.
The system follows the law of each country.
Rarus Health designs the system to meet Chile's Ley 19.628 and Ley 21.719, and the applicable law in each country where it works.
Read about data protectionHow does Rarus Health choose its technology?
Rarus Health builds on proven, widely used technology. Four rules guide each choice, so the system stays easy to maintain as the technology changes.
Models are replaceable.
Rarus Health uses language, vision and speech models from leading providers. Rarus Health chooses a model for each task and changes it when a better one appears. The rest of the system does not change.
Channels are official.
Rarus Health reaches people through the official WhatsApp Business Platform, with delivery tracking for every message.
The core is standard.
The record runs on standard, open technologies that many engineers know. Rarus Health avoids tools that would lock the company to one supplier.
The infrastructure is defined as code.
Rarus Health runs on managed Google Cloud services. Code defines every environment, so Rarus Health can rebuild or audit any environment from that code.
Talk to Rarus Health
Ask for a demonstration of the service on WhatsApp, or for a technical briefing.
Identify where patients drop outDo you want the business view first? Read the How it works page. Do you want to know who builds the technology? Meet the team.