Hekavor
Taking the Patient History With AI, Before the Visit Begins
.jpg)
AI Patient Intake
Working Differential
Chart-Ready Notes
Patient & Provider Apps
HIPAA Compliance
Mari is Hekavor’s AI medical assistant. It runs a physician-level patient history before the visit begins, builds a working differential, and hands the doctor a chart-ready note. Everything that follows rests on that detail.
Topflight took Hekavor’s working prototype to production across two apps, a patient intake and a provider portal, on a HIPAA-compliant backend, in time for the founder’s first conference.
By the Numbers
2 apps
A patient intake and a provider portal in one system
On time
Delivered for the founder’s first conference
5.0
Clutch review rating across quality and schedule
~80%
Of a diagnosis rests on the patient history Mari automates
Where Mari Fits
.jpg)
Building Primary Care’s Pre-Encounter Layer
Hekavor is developing AI-assisted tools for primary care. Founded by Adrit Lath and Dr. Sara Rotner in Mountain View, California, the company is building the pre-encounter clinical intelligence layer. Its assistant, Mari, handles the opening conversation of a visit.
Asks what is wrong and what else is going on
Returns the first ten minutes of each appointment
Feeds diagnosis decision support
Validated on hundreds of physician-developed cases
Adrit had already built a working prototype for specific chief complaints. To reach the general population, he needed a partner who had shipped HIPAA-compliant healthcare apps before.
Four Problems Between Prototype and Production
01
Productionize a Working Prototype
Move from a prototype built for a handful of chief complaints to a production-grade app ready for the general population.
02
HIPAA With LLMs in the Loop
Patient data flows to and from large language models, which meant business associate agreements and secure data handling.
03
Two Apps, One System
A patient-facing intake named Mari and a provider portal, both built around the core LLM logic that Hekavor owns and manages.
04
Groundwork for EHR Integration
Lay the technical and compliance foundation for integrating with major systems like Epic and Cerner, central to the long-term plan.
Solution: From Patient Intake to a Chart-Ready Note
Six pieces move a patient from the first question to a note the physician can chart, across two apps.
01
Mari AI Patient Intake
A physician-level history across a widening set of chief complaints, asked as grouped multiple-choice questions.
02
Provider Portal
Providers manage patients, schedule Mari links, review summaries, and track usage from one back office.
03
Working Differential and Note
Mari mirrors how physicians document: present illness, review of systems, past history, a differential, and next steps.
04
Roles and Permissions
Distinct roles for providers and assistants, with granular control over links, records, and case details.
05
Customizable Notifications
Per-patient control over Mari email content and notification channels, across both SMS and email.
06
Usage Tracking and Error Logs
An admin view into system health, user interactions, and technical errors, so issues surface before users feel them.
Under the Hood
Engineering an LLM Intake That Doesn’t Skip a Question
Challenge
The core IP, the physician-developed question logic, stays with Hekavor. Mari runs a hybrid design: deterministic question pathways the conversation follows reliably, with a large language model layered on for natural-language flexibility.
.jpg)
Solution
Defensive handling catches skipped questions and hallucinated facts.
Prompt adjustments keep the intake predictable from first question to last.
Output stays structured cleanly enough to extract into a physician’s record.
1
Patient Intake
A physician-level history, one grouped question at a time.
2
Defensive Handling
Patches catch skipped questions and hallucinated facts.
3
HIPAA-Compliant LLM Pass
Patient data moves through Azure OpenAI under a BAA.
4
Structured Note Output
A chart-ready note with a working differential and next steps.
Running an LLM on Patient Data, the Compliant Way
Patient data moves to and from an LLM here, so every tool in the stack had to handle PHI under a BAA and still let a small team ship fast.
React
Patient and provider web apps
FastAPI
Backend API
Azure OpenAI
LLM engine, BAA in place
Supabase
HIPAA-compliant Postgres
Twilio + SendGrid
Patient SMS and email
Render
Dev, staging, production
On Time for the Conference
Hekavor is an early-stage company, so the early wins matter: a clean launch, a first set of interested users, and a partnership that kept going.
On-time delivery
Conference-ready
The production app was finished in time for the founder’s first conference, and the demo’s opening day ran flawlessly.
Pipeline started
~10 prospects
Around ten people asked to learn more before signing up, with trial calls scheduled over the following weeks.
Ongoing partnership
Phase 2
The engagement moved into a maintenance phase and a planned Phase 2, after the team hit a deadline that triggered a delivery bonus.
5.0
Quality
5.0
Schedule
4.5
Cost
5.0
Willing to Refer
Turning an AI Prototype Into a Compliant Product?
We turn healthcare AI prototypes into production-grade, compliant apps, from your core IP to launch, in a single quarter or two.