Hospitalflow AI

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle

What I Built

HospitalFlow AI is a web dashboard that spots operational bottlenecks in hospitals before they cause delays. It compares each department’s workload (appointments, task time, emergency cases) with its staff capacity and flags it as Normal, Warning or Critical. A local AI model then explains the problem and suggests three short management actions.

Features

Live dashboard with risk counts and color-coded departments
AI Bottleneck Detector with evidence, explanation and suggested actions
Tomorrow’s forecast based on the last 7 days
“Ask the Data” box for plain-language questions (medical questions are refused)

Demo

https://drive.google.com/drive/my-drive?dmr=1&ec=wgc-drive-%5Bmodule%5D-goto

Code

How I Built It

I built an offline hospital bottleneck dashboard using Python and FastAPI, with a plain HTML/CSS/JS frontend served by the same app.

Python computes every number (workload vs staff capacity, risk levels, tomorrow’s forecast) from synthetic, non-clinical data. A local Gemma model, run through Ollama, only explains the results and suggests actions, and rule-based fallback keeps the app working if the AI is unavailable.

I wrote one detailed prompt with strict rules and used the Antigravity IDE agent to build, test and push the code to GitHub, then deployed it on Render.

Why Does Open Innovation Matter?

Open innovation means building on shared tools, models and ideas instead of starting from scratch, and sharing back what we learn. It matters because it lets small teams build things that once needed big budgets.

My project is an example. HospitalFlow AI runs on open-source pieces: Python, FastAPI and pandas for the app, plus an open-weights model, Gemma, served locally by Ollama. Because these are open, I could build a private, offline hospital tool in a hackathon, with no paid API and no patient data leaving the machine.

Open work also improves faster, since many people test it, fix it and extend it. It lowers the barrier for students and small communities, and it keeps control with the people who use the tools, which matters for privacy-sensitive fields like healthcare.

Takeaway: open tools let anyone build something useful, so sharing code, models and lessons helps everyone move faster.

My Agent Session

I used the Antigravity IDE agent to build HospitalFlow AI. I gave it one detailed prompt with the architecture, hard rules (synthetic data only, no cloud AI, AI never does math), file structure and a testing checklist.

The agent planned the work, wrote the FastAPI backend and the HTML/CSS/JS frontend, ran the server, tested the endpoints, and pushed the code to GitHub. I then deployed it on Render.

Lesson: an agent works best when you give it clear rules and ask it to verify its own output.

My First User

My first user was [name or role, e.g. a classmate, a friend, a teammate], who opened HospitalFlow AI at https://hospital-workflow-g8rq.onrender.com/ and tried it without any help from me.

What they did first: [e.g. clicked straight to the red Radiology alert / asked “Which departments are critical?”].

What I learned: [e.g. the first load was slow because the free server was asleep / they wanted the banner to explain “fallback mode” / the example questions made it easy to start].

What I’m changing because of it: [e.g. add a short “what is this?” line at the top, or a loading message while the server wakes up].

Takeaway: watching one real person use your project shows problems you can’t see as the builder.