PulseShield

Inspiration

Most safety systems can detect an abnormal sensor reading, but they rarely understand the bigger picture. A temperature spike alone does not explain whether a building is experiencing an electrical fault, fire, equipment failure, or a harmless anomaly.

We wanted to build something closer to a safety intelligence system—one that can combine multiple signals, understand what is happening, decide what should happen next, and communicate that response clearly.

That idea became PulseShield: a software-based cyber-physical safety platform that uses simulated IoT sensors, AI reasoning, deterministic safety logic, and a real-time digital twin to demonstrate how intelligent buildings could detect and respond to dangerous situations.

What it does

PulseShield continuously simulates a building's physical environment through virtual sensors and represents it through a living digital twin.

It can:

• Monitor temperature, smoke, gas, current, voltage, vibration, water, occupancy, air quality, doors, network and battery status.
• Generate realistic real-time telemetry instead of static dashboard values.
• Combine multiple sensor signals to identify potential hazards.
• Detect scenarios such as electrical overheating, fire development, gas leaks, water leaks, sensor failures, network failures and multi-zone emergencies.
• Classify incidents by severity and confidence.
• Use a deterministic safety engine to control critical safety decisions.
• Simulate physical responses such as circuit isolation, alarms, emergency lighting, HVAC shutdown and zone isolation.
• Calculate safer evacuation routes based on changing hazards and occupancy.
• Display the entire building state through an interactive digital twin.
• Record incidents as chronological timelines.
• Replay incidents and synchronize historical sensor, building and actuator states.
• Provide an AI Safety Analyst that explains what happened and why.
• Generate structured AI-powered incident reports and maintenance recommendations.
• Monitor sensor health and detect degraded or failed sensors.
• Continue core safety behavior in simulated offline/edge conditions.
• Provide a dedicated Demo Mode for repeatable hackathon demonstrations.

The key idea is:

Observe → Understand → Decide → Act → Guide → Learn

How we built it

PulseShield was designed as a cyber-physical architecture even though our prototype does not require physical hardware.

The architecture is:

Virtual Sensors → Real-Time Simulation → Sensor Fusion → AI Hazard Intelligence → Deterministic Safety Engine → Virtual Actuators → Digital Twin → Incident Replay & Analytics

We built the application using a modern web stack including Next.js, React, TypeScript, Tailwind CSS and Framer Motion, with real-time telemetry, interactive visualizations and charting.

The virtual hardware layer simulates an edge gateway and multiple IoT sensors. A scenario engine changes sensor behavior over time to reproduce realistic emergency conditions.

A sensor-fusion layer evaluates multiple signals together instead of relying on a single reading.

AI is used primarily for interpretation, explanation, incident analysis and recommendations. Critical safety decisions remain controlled by deterministic rules, making the architecture safer and more predictable than allowing an LLM to directly control emergency actions.

The digital twin maps telemetry and safety state back onto individual building zones, while the incident system records every important event for later replay and analysis.

The simulator is intentionally designed so that real ESP32/STM32 sensors could replace the virtual sensor layer in a future physical deployment without redesigning the core safety architecture.

Challenges we ran into

One of the biggest challenges was avoiding the typical "IoT dashboard" approach.

Simply displaying sensor values was not enough. We needed to create a complete chain from a changing physical environment to intelligent interpretation and a visible response.

Another challenge was designing AI responsibly. We did not want an LLM making unrestricted emergency decisions. We separated AI reasoning from deterministic safety control, allowing AI to explain and analyze situations while the safety engine controls critical actions.

We also had to make simulated hardware feel realistic. Static values would make the project look artificial, so we introduced changing telemetry, sensor reliability, abnormal trends, failures, conflicting signals and multi-zone scenarios.

Finally, we had to synchronize many parts of the system—telemetry, digital twin states, incidents, evacuation routes, actuators and replay—while keeping the interface responsive and understandable during emergencies.

Accomplishments that we're proud of

We are proud that PulseShield goes beyond detection and demonstrates an entire cyber-physical response loop in software.

Our biggest accomplishments include:

• Building a real-time virtual IoT environment without requiring physical hardware.
• Creating a living digital twin that reacts to simulated physical conditions.
• Implementing multi-sensor hazard reasoning instead of single-threshold alerts.
• Separating AI interpretation from deterministic safety decisions.
• Creating automated virtual safety responses.
• Building dynamic evacuation intelligence based on hazards and occupancy.
• Creating synchronized incident timelines and historical replay.
• Adding an AI Safety Analyst that can explain system decisions.
• Supporting sensor and network failure scenarios.
• Creating a repeatable emergency simulation for live demonstrations.
• Designing the experience as a premium safety command center rather than a conventional analytics dashboard.

Most importantly, we built PulseShield around a simple principle:

Don't just detect danger. Understand it. Respond to it. Protect people.

What we learned

We learned that building an intelligent system is not only about adding AI.

The real engineering challenge is deciding where AI should be used and where deterministic logic should remain in control.

We also learned the importance of sensor fusion. A single abnormal measurement can be misleading, while multiple correlated signals can provide much stronger evidence.

Building the digital twin taught us that visualization can become part of the reasoning experience—not just decoration. When the building, telemetry, incidents and evacuation routes change together, complex system behavior becomes much easier to understand.

We also learned that failure scenarios are just as important as successful scenarios. A system that works only when every sensor, network and AI service is healthy is not a resilient safety system.

What's next for PulseShield

The next step is to move PulseShield from a software-based digital twin into a real hardware-in-the-loop platform.

We plan to connect:

• ESP32/STM32 edge devices
• Real temperature, smoke, gas, current and vibration sensors
• Real-time MQTT/industrial IoT communication
• Edge ML for local anomaly detection
• Thermal/RGB cameras for visual verification
• Physical relays and emergency indicators
• Robotic inspection for high-risk areas
• Multi-building monitoring
• Predictive maintenance
• Historical safety analytics
• Stronger offline/edge autonomy

The long-term vision is to make PulseShield a hardware-agnostic safety intelligence layer that can sit between physical infrastructure and the people responsible for protecting it.

A future PulseShield deployment would continuously observe a real environment, understand emerging risks, take predefined safe actions, guide occupants, and create an auditable record of what happened.

From simulated sensors today to intelligent physical safety infrastructure tomorrow.

Built With

• python