MedOrbit

MedOrbit homepage.

Inspiration

In today’s healthcare systems, conversations between doctors and patients carry critical insights—but most of that information is lost, unstructured, or manually documented after the fact.

We were inspired by a simple idea:
What if conversations themselves could become structured clinical intelligence in real time?

At the same time, there is a growing fear that AI might replace professionals. We wanted to flip that narrative—AI should assist and empower doctors, not replace them. MedOrbit is our attempt to bridge that gap.

What It Does

MedOrbit is an AI-assisted clinical and behavioral intelligence platform that transforms live doctor–patient conversations into actionable insights.

With a single interaction (clicking the orb), MedOrbit:

Captures consultation conversations
Generates real-time transcript streams
Extracts behavioral and clinical signals
Produces structured summaries and draft reports
Enables doctor review and approval
Delivers simplified explanations and reminders to patients.

The system ensures a strict boundary:

Patients only see doctor-approved content, never raw or unverified AI output.

How We Built It

We designed MedOrbit as a modular, multi-agent system:

AI Architecture

Behavioral Agent → detects emotional and behavioral signals
Triage Agent → extracts symptoms and clinical context
Super Agent (Orchestrator) → combines outputs into structured reports

System Architecture

Frontend: React + TypeScript (custom UI with animated interaction layer)
Backend: FastAPI with role-based APIs
Database: PostgreSQL (visits, reports, reminders)
Auth: JWT-based authentication with doctor/patient roles

UX Innovation

Our key differentiator is the blob interaction:

Users click an animated orb on the landing page.
It expands into a live consultation workspace.
Transcript and AI insights appear progressively.

This turns a complex AI pipeline into one intuitive interaction.

Challenges We Faced

1. Real-time conversation processing

Handling live transcription and speaker identification is complex.
Solution: For the MVP, we simulated real-time transcript streaming while keeping the architecture compatible with real audio pipelines.

2. Avoiding “AI hallucination” risk

Healthcare requires strict reliability.
Solution: We introduced a doctor-in-the-loop approval system, ensuring:

AI outputs are drafts.
Only approved content reaches patients.

3. UX vs complexity trade-off

We had a powerful backend but needed a simple, intuitive interface. Solution: The orb interaction became the central entry point, abstracting complexity into a seamless flow.

4. End-to-end integration under time constraints

Connecting agents, backend APIs, UI, and demo flow within limited time was challenging.
Solution: We used a demo-safe architecture with deterministic data flows to ensure reliability.

What We Learned

Designing AI systems is not just about models—it’s about **trust, flow, and usability.
Real-world AI products must include **human oversight loops.
Strong UX can make complex systems feel simple and intuitive.
Building for a hackathon requires balancing innovation with reliability