Real-time NICU clinical decision support with human-in-the-loop sepsis risk management.
Neovance AI is a full-stack platform for early neonatal sepsis risk detection in NICU environments. The system combines live vital signals, rule-based clinical context, and machine learning inference to support high-quality doctor decisions without replacing clinical authority.
The core product philosophy is human-in-the-loop (HIL): AI identifies risk, clinicians make the final decision, and those outcomes are captured to continuously improve future model behavior.
The initial prototype was built during Singularity, a 30-hour AI hackathon at IIT Dharwad, and has since been organized into a submission-ready engineering repository.
| Layer | Technologies | Purpose |
|---|---|---|
| Frontend | Next.js, React, TypeScript, Tailwind CSS, shadcn/ui, Chart.js | Role-based dashboards for doctors and nurses, live trend visualizations |
| Backend API | FastAPI, Uvicorn, WebSockets | Real-time data APIs, alert workflows, live updates |
| Data and Persistence | SQLite (current runtime), PostgreSQL/TimescaleDB (architecture target) | Clinical records, vitals history, HIL action logs |
| ML and Analytics | scikit-learn, joblib, pandas, numpy | Sepsis risk scoring and model artifact loading |
| Streaming and Simulation | Pathway pipeline modules, vitals simulators | Streaming ingestion and realistic patient signal generation |
| Developer Tooling | Python venv, npm, ESLint | Local development, static checks, reproducible setup |
- Real-time neonatal vitals monitoring with continuously updated dashboards.
- ML-powered sepsis risk prediction and alert generation.
- Human-in-the-loop doctor action panel (observe, tests, treat, dismiss).
- Role-based workflow for doctor and nurse responsibilities.
- Alert auditability and decision traceability for clinical accountability.
- EOS-oriented workflow components and simulator support.
- WebSocket-based live feed for low-latency UI updates.
- Vitals and clinical context are ingested from simulators or streaming sources.
- Backend computes clinical features and invokes model inference.
- If risk crosses threshold, an alert is created and surfaced to doctors.
- Doctor action is recorded (observe/tests/treatment/dismiss).
- Nurse receives instructions and continues bedside monitoring.
- Final outcomes are logged for performance review and future retraining loops.
- Python 3.9+
- Node.js 18+
- npm
- Optional: Docker (for PostgreSQL/TimescaleDB setup)
git clone https://github.com/jonahmichael/Neovance-AI.git
cd Neovance-AIWindows (PowerShell):
python -m venv venv
.\venv\Scripts\Activate.ps1
pip install -r requirements.txtLinux/macOS:
python -m venv venv
source venv/bin/activate
pip install -r requirements.txtcd frontend/dashboard
npm install
cd ../..If trained model files are missing, generate them:
python scripts/train_sepsis_model.pycd backend
python -m uvicorn main:app --reload --host 0.0.0.0 --port 8000cd frontend/dashboard
npm run devpython scripts/run_ml_hil_system.py
python scripts/run_neovance.py- Frontend: http://localhost:3000
- Backend API: http://localhost:8000
- Backend API Docs: http://localhost:8000/docs
Credentials are documented in docs/credentials.md.
Neovance-AI/
|-- backend/ # FastAPI services, data models, simulators, ETL
|-- frontend/dashboard/ # Next.js UI and role-based clinical dashboards
|-- scripts/ # Training, orchestration, and workflow utilities
|-- docs/ # Setup, API, deployment, and feature documentation
|-- data/ # Stream and simulation input files
|-- tests/ # System-level tests
`-- requirements.txt # Python dependency manifest
This repository is organized to support submission review in a clear, production-oriented format.
- End-to-end NICU monitoring flow from live vitals to clinician decision logging.
- Real-time dashboard and API surface for demos and validation.
- Layered structure by concern (backend, frontend, scripts, docs, tests).
- Typed frontend codebase and lint-ready JavaScript/TypeScript tooling.
- Modular backend components and documented APIs.
- Event/stream processing architecture prepared for high-frequency vital inputs.
- Decoupled UI/API/ML components enable independent scaling and deployment.
- Human-in-the-loop feedback loop captures clinician actions as high-value training signals.
- Integration of live monitoring, risk prediction, and role-aware clinical workflows.
- Setup guide: docs/setup.md
- Feature specification: docs/features.md
- API reference: docs/api.md
- Deployment guide: docs/deployment.md