Neural Zero ❤️

I built Neural Zero as a predictive surveillance system for patients with cardiovascular history who have a hard time communicating and expressing themselves.

Instead of waiting for a patient to fully crash and then react, I wanted to show how a room camera, a small biometric pipeline, and a multi-agent console could help flag deterioration earlier.

Next.js React TypeScript Claude Anthropic API Vercel

Live demo on Vercel

What this is ✨

Neural Zero is a hackathon prototype and demo experience powered by six Claude agents.

I designed it around one simple idea:

If a patient cannot speak or move much, their face, breathing pattern, color return, and tiny physiological changes may still be telling us something important before a classic bedside alarm fires.

So I built a product story around that.

The project has:

• a landing page that explains the problem and the system
• a patient directory
• a dossier page with human context, baseline notes, and clinician-style context
• a war room where six Claude agents debate what the system is seeing
• a report view that summarizes technical findings, clinical interpretation, and final action

The demo flow 🎥

This is the flow I use when I present it:

1. I start on the homepage.
2. I open the patient directory.
3. I search for a patient and open their dossier.
4. I enter the war room.
5. The center canvas shows the patient relay.
6. The agents around the edges explain what each lane is seeing.
7. The Chief terminal resolves the room into one final decision.

I currently use two demo patients:

• PT-001 / Morgan Ellison: quiet baseline, observe-only case
• PT-002 / Rowan Hale: distress case, predictive escalation case

What I wanted to prove 🎯

I wanted this project to feel like something that could sit between messy real hospital infrastructure and an actual clinical team, with six Claude agents doing the reasoning work.

That is why I included:

• a hidden cached/live relay mode
• a historical nearest-case graph lane
• a Holter-style electrical comparator for the Bio lane
• a glassy but readable console that feels more like a workstation than a generic dashboard

Vision note 👀

The Vision lane is built to work with a real camera feed, but for the demo I layered the visual cues, scan overlays, and face locks so the experience is still believable and easy to present without a live camera setup.

Main features 🧠

1. Landing page 🌐

I wanted the first impression to feel calm, technical, and intentional.

The homepage explains:

• the clinical problem
• the predictive-monitoring angle
• the multi-agent reasoning model
• why silent or preventable deterioration matters

2. Patient dossier 📄

I did not want the patient pages to feel like anonymous machine output.

So each dossier includes:

• human background
• medical history
• hospital stay context
• physician note
• nursing note
• baseline persona based on time already spent in the hospital

3. War room ❤️

The war room is the core experience.

It includes:

• live patient feed
• ECG strip
• agent cards around the edges
• a Vision popup with face lock, body scan, and cue overlays
• a CrossCheck popup with a knowledge-graph style view
• a Bio popup with a Holter deterioration replay
• a Chief consensus stream
• a downloadable clinical report
• six Claude agents, each with a distinct role and voice

Tech stack ⚙️

• Next.js 16
• React 19
• TypeScript
• Tailwind CSS
• Radix UI
• Lucide React
• Claude via the Anthropic API for the live relay mode

Project structure 🗂️

app/
  api/
  console/
components/
  console/
  landing/
  ui/
lib/
scripts/
src/
  dataset/
  imgs/
  vids/

Local setup 🚀

1. Install dependencies

pnpm install

2. Add environment variables

Create .env.local with:

ANTHROPIC_API_KEY=your_key_here
ANTHROPIC_MODEL=claude-sonnet-4-20250514

3. Run the app

pnpm dev

4. Open it

http://localhost:3000

If I am using a different port locally, I just open that port instead.

Live mode vs cached mode 🔄

I built the console so it can work in two ways:

• cached: best for demos, stable and predictable
• live: pulls server-side generated war-room output through Claude via the Anthropic API

If the API key is missing, the app still works in cached mode.

About the dataset folder 📁

The files inside src/dataset are there for presentation and context.

They are not required for the main app flow, and I did not treat them as a production-grade medical backend.

Why I made it this way 💭

I did not want this to feel like a generic "AI healthcare dashboard."

I wanted it to feel like:

• a real pitch
• a believable clinical command center
• a system with layered reasoning
• a product demo that tells a story clearly

So the whole repo is built around clarity, atmosphere, and believable clinical storytelling.

Important note ⚠️

This is a prototype for demonstration purposes, not a real medical device, not a diagnostic tool, and not clinical advice. Also, the patient videos are AI Generated !

Author 👋

Built by me, Aashir.

If you are here from the demo, thank you for taking the time to look through it.