MediOS — I Building an AI Hospital Control Plane Inside Notion

Shridhar Akki — Sun, 22 Mar 2026 09:22:11 +0000

This is a submission for the Notion MCP Challenge

Video Demo

watch in youtube
https://youtu.be/xp2dmOBo06U

The Problem

Hospitals don’t struggle because of lack of expertise —

they struggle because of everyday workflow friction.

Healthcare work is inherently hectic.

Doctors and nurses operate under constant time pressure,
handling multiple patients, decisions, and updates at once.

Supplies need to be tracked manually, and low stock is often noticed too late. masks , medicines , oxygen cylinder much more etc
Patient information is recorded in multiple places, making it hard to keep everything consistent,The test records various department,patient history
Doctor notes are sometimes difficult to read or follow, especially under time pressure These tasks are routine, but they take time, require coordination, and are easy to miss.

I wanted to explore:

What if these routine tasks could become simple, structured, and worry-free —

so hospital staff can focus on decisions instead of coordination?

People get tired and exhausts because of paperwork and disconnected systems.

What I Built

MediOS — an AI Clinical Control Plane that triages patients,
tracks hospital supplies, and keeps humans in control of every
critical decision.

The core idea: AI handles the paperwork. Humans make the decisions.

Show us the code

GitHub: https://github.com/shridharakki/medios

System Architecture

Nurse / Doctor types a message
           ↓
    Claude Desktop
           ↓
  Custom MCP Server (server.py)
           ↓
      5 MCP Tools:
  triage_patient()        → scores symptoms 0-100
  update_inventory()      → updates stock levels
  process_approved()      → assigns nurse on approval
  update_patient_notes()  → updates prescription
  create_alert()          → fires automatically on critical stock
           ↓
      Notion API
           ↓
  3 Databases: Patients | Review Queue | Inventory
           ↓
  watcher.py (background — checks every 30 seconds)
  → detects doctor approvals
  → assigns nurses automatically
  → zero extra commands needed

How I Used Notion MCP

Notion is not just a database in MediOS — it is the decision
surface where humans and AI collaborate.

3 Databases

Patients DB
Stores every patient with: Name, Symptoms, AI Score (0-100),
Priority (Critical/High/Medium), Status, Assigned Nurse, Doctor Notes

Review Queue DB
Every AI decision goes here first — nothing happens without
human approval. Doctor sees the AI reasoning and approves or rejects.

Inventory DB
Real-time supply tracking — Quantity, Threshold, Status
(OK/LOW/CRITICAL), Needs Reorder checkbox

3 Demo Flows

Flow 1 — Patient Triage + Human Approval

Nurse types:

"Add patient Ravi, chest pain and shortness of breath"

AI automatically:

Scores symptoms → 91/100 — Critical
Creates entry in Patients DB
Sends to Review Queue as Pending

Doctor opens Notion → reads AI reasoning → changes to Approved

Watcher detects approval → Nurse Priya assigned instantly

No extra command. No manual step. The system reacts on its own.

Flow 2 — Inventory Alert (Proactive AI)

Nurse types:

"Update gloves quantity to 5"

AI automatically:

Updates Inventory DB
Detects quantity (5) is below threshold (20)
Sets status → CRITICAL
Ticks Needs Reorder checkbox
Creates management alert in Review Queue

**One message. Five actions.

This is the difference between automation and intelligence —
the AI noticed a problem and reported it without being asked.

Flow 3 — Doctor Prescription

Doctor types:

"Update Ravi's notes: Prescribe aspirin 75mg, follow up in 3 days"

AI automatically:

Updates Doctor Notes in Patients DB
Changes patient Status → Treated

Doctor focuses on medicine. AI handles the paperwork.

Tech Stack

Component	Technology
AI Brain	Claude Desktop
MCP Server	Custom Python (server.py)
Automation	Python watcher script (watcher.py)
Database	Notion (3 databases)
Packages	notion-client, python-dotenv, mcp
Language	Python 3.13

Future Roadmap

MediOS today handles triage, inventory, and prescriptions.
But the vision is much bigger —
There so many features to add
A complete Hospital Operating System where every role,
every department, and every patient has their own
intelligent workflow.

Phase 1 — Multi-User & Custom Tools

Role-based access — Doctor, Nurse, Patient, Management each get their own custom MCP tools
Doctor tools: voice dictation, prescription writer, patient summary
Nurse tools: inventory reporting, patient updates, task management
Management tools: dashboards, approvals, alerts
Patient tools: view own records, book appointments, ask health questions

Phase 2 — Full Hospital Departments

Medicine & Supplies — automated reordering, vendor alerts, usage prediction
Instruments & Equipment — maintenance schedules, availability tracking
Examinations & Tests — lab report scanning, result auto-updates to patient record
Radiology & Scans — scan booking, report summarisation, doctor review queue
Reporting — daily department summaries, weekly management reports auto-generated
Appointments & Booking — patient self-booking, doctor calendar sync, reminder alerts

Phase 3 — Patient Home Care

Medicine reminders — WhatsApp/SMS alerts so patients never forget their medication
Personal health assistant — patient asks simple questions like:
- "What should I eat today?"
- "Which exercises help my recovery?"
- "What activities should I avoid?"
- "What yoga is good for my condition?"
AI answers based on their actual prescription and doctor notes — not generic internet advice
Recovery schedule — personalised daily plan sent to patient automatically after discharge
Follow-up booking — AI reminds patient when next appointment is due and books it automatically

Phase 4 — Hospital Intelligence

AI predicts busy periods from historical patient data
Real-time bed availability across all wards
Staff scheduling suggestions based on patient load
Emergency ambulance tracking integration
Insurance claim automation
Cross-department coordination — lab results automatically notify the right doctor instantly

The Big Picture

From the moment a patient enters the hospital
to the moment they recover at home —
every step tracked, every task automated,
every human decision supported by AI.

MediOS is not just a hospital tool.
It becomes the operating system for healthcare.

What I Learned

This was my first real coding project outside of college assignments.

I learned:

How to build custom MCP and how to connect it.
How MCP servers work and how to build one from scratch
How to connect Python to external APIs
How to think about system design — not just writing code

DEV Community: Shridhar Akki

MediOS — I Building an AI Hospital Control Plane Inside Notion

Video Demo

The Problem

What I Built

Show us the code

System Architecture

How I Used Notion MCP

3 Databases

3 Demo Flows

Flow 1 — Patient Triage + Human Approval

Flow 2 — Inventory Alert (Proactive AI)

Flow 3 — Doctor Prescription

Tech Stack

Future Roadmap

Phase 1 — Multi-User & Custom Tools

Phase 2 — Full Hospital Departments

Phase 3 — Patient Home Care

Phase 4 — Hospital Intelligence

The Big Picture

What I Learned