Category
AI-Empowered IOT Device
Time
Feb - May 2025
Team
Individual Project
Tools
Python, Adafruit IO, ProtoPie, Bosch AI Studio, Figma
A wildfire detection ecosystem bridging the gap between delayed official alerts and on-the-ground reality. It verifies local threats in real-time to create a trusted safety network. Operating as a sustainable social enterprise, hardware revenue funds the platform, ensuring life-saving intelligence remains unconditionally free for every resident.
A wildfire detection ecosystem bridging the gap between delayed official alerts and on-the-ground reality. It verifies local threats in real-time to create a trusted safety network. Operating as a sustainable social enterprise, hardware revenue funds the platform, ensuring life-saving intelligence remains unconditionally free for every resident.
A wildfire detection ecosystem bridging the gap between delayed official alerts and on-the-ground reality. It verifies local threats in real-time to create a trusted safety network. Operating as a sustainable social enterprise, hardware revenue funds the platform, ensuring life-saving intelligence remains unconditionally free for every resident.
Research
Research
In early January 2025, a series of wildfires struck the Los Angeles area of California, driven by powerful Santa Ana winds; as the Palisades, Eaton Fires and the Hurst Fire burned across the county, the disaster resulted in 31 deaths, forced more than 200,000 people to evacuate, and destroyed over 18,000 structures.
In early January 2025, a series of wildfires struck the Los Angeles area of California, driven by powerful Santa Ana winds; as the Palisades, Eaton Fires and the Hurst Fire burned across the county, the disaster resulted in 31 deaths, forced more than 200,000 people to evacuate, and destroyed over 18,000 structures.
In early January 2025, a series of wildfires struck the Los Angeles area of California, driven by powerful Santa Ana winds; as the Palisades, Eaton Fires and the Hurst Fire burned across the county, the disaster resulted in 31 deaths, forced more than 200,000 people to evacuate, and destroyed over 18,000 structures.
31
Death
Death
Death
250,000 +
People Evacuated
People Evacuated
People Evacuated
18,000 +
Structure Destoryed
Structure Destoryed
Structure Destoryed
The disconnect expressed by residents prompted a deep dive into the systemic friction within the current warning process…
The disconnect expressed by residents prompted a deep dive into the systemic friction within the current warning process…
The disconnect expressed by residents prompted a deep dive into the systemic friction within the current warning process…
Problems
Problems
Current wildfire warning procedures are fragmented, causing delays in community awareness and slowing evacuation response
Current wildfire warning procedures are fragmented, causing delays in community awareness and slowing evacuation response
Current wildfire warning procedures are fragmented, causing delays in community awareness and slowing evacuation response
Primary Research
Primary Research
To gain a deep understanding of the confusion and anxiety residents experience when navigating fragmented warning systems.
To gain a deep understanding of the confusion and anxiety residents experience when navigating fragmented warning systems.
To gain a deep understanding of the confusion and anxiety residents experience when navigating fragmented warning systems.
Interview
Interview
5
5
Residents
Who were affected by the Eaton Fire in Altadena Area
Interview insights
Interview insights
Problem Statements
Current wildfire information and alerts are scattered across many platforms, and residents struggle to understand and trust this information
Current wildfire information and alerts are scattered across many platforms, and residents struggle to understand and trust this information
Opportunities
A community-based wildfire detection ecosystem for local residents that unifies data and alerts into one trusted, transparent platform.
A community-based wildfire detection ecosystem for local residents that unifies data and alerts into one trusted, transparent platform.
Solution
Solution
Community Wildfire Detection Ecosystem
Community Wildfire Detection Ecosystem
Community Wildfire Detection Ecosystem
Our solution is a community ecosystem that unifies diverse data sources into a single, transparent public platform.
Our solution is a community ecosystem that unifies diverse data sources into a single, transparent public platform.
Our solution is a community ecosystem that unifies diverse data sources into a single, transparent public platform.
Hardware Prototype
Hardware Prototype
Product Definition
To bridge the trust gap between official alerts and scattered data sources, this device provides transparent, first-hand verification of fire threats directly from the community
To bridge the trust gap between official alerts and scattered data sources, this device provides transparent, first-hand verification of fire threats directly from the community
To bridge the trust gap between official alerts and scattered data sources, this device provides transparent, first-hand verification of fire threats directly from the community
Detect and Analyze Wildfire
Combines multiple sensors to detect early fire signals, run basic analysis on-device, and able to send clean data to the cloud.
Combines multiple sensors to detect early fire signals, run basic analysis on-device, and able to send clean data to the cloud.
Sustainable
Designed for autonomy, utilizing efficient solar panels and ruggedized casing to operate continuously in harsh outdoor environments without maintenance.
Designed for autonomy, utilizing efficient solar panels and ruggedized casing to operate continuously in harsh outdoor environments without maintenance.
Product Design Criteria
Product Design Criteria
Compact & Unobtrusive
Compact & Unobtrusive
Compact & Unobtrusive
Self-Sustaining
Self-Sustaining
Self-Sustaining
Weather Resistant (IP67)
Weather Resistant (IP67)
Weather Resistant (IP67)
Good Sensor Exposure
Good Sensor Exposure
Good Sensor Exposure
FireGuard Device
A resilient, solar-powered sensing node that uses multi-sensor fusion and edge computing to validate fire threats in real-time
A resilient, solar-powered sensing node that uses multi-sensor fusion and edge computing to validate fire threats in real-time
A resilient, solar-powered sensing node that uses multi-sensor fusion and edge computing to validate fire threats in real-time
Hardware architecture schematic and physical prototype for testing real-time environmental sensing
Hardware architecture schematic and physical prototype for testing real-time environmental sensing
Final 3D Printed Functional Prototype
Final 3D Printed Functional Prototype
Software Prototype
Software Prototype
Cloud Connectivity
Cloud Connectivity
Fire Detection
Fire Detection
Self-Sustaining
Self-Sustaining
AI Classification
AI Classification
Research indicates that static threshold-based rules often yield high false-positive rates due to environmental noise (e.g., dust or humidity spikes) [1]
Research indicates that static threshold-based rules often yield high false-positive rates due to environmental noise (e.g., dust or humidity spikes) [1]
Research indicates that static threshold-based rules often yield high false-positive rates due to environmental noise (e.g., dust or humidity spikes) [1]
To address this, I leveraged the embedded Edge-AI capabilities of the Bosch BME sensor. Instead of relying on cloud processing, the sensor itself was trained to recognize distinct 'gas signatures'—distinguishing the specific chemical composition of smoke from benign natural interferences locally.
To address this, I leveraged the embedded Edge-AI capabilities of the Bosch BME sensor. Instead of relying on cloud processing, the sensor itself was trained to recognize distinct 'gas signatures'—distinguishing the specific chemical composition of smoke from benign natural interferences locally.
To address this, I leveraged the embedded Edge-AI capabilities of the Bosch BME sensor. Instead of relying on cloud processing, the sensor itself was trained to recognize distinct 'gas signatures'—distinguishing the specific chemical composition of smoke from benign natural interferences locally.
99.21%
99.21%
Accuracy in classifying smoke and fire using AI
Accuracy in classifying smoke and fire using AI
While the AI classifier achieved 99.21% accuracy in validation, this serves as a technical exploration. Real-world deployment would require significantly larger datasets to account for diverse fire types and outdoor variability.
Training Data used from PurpleAir.com
[1] Sowah, R., et al. (2016). "Design of a Fire Detection and Control System using Fuzzy Logic."
[1] Sowah, R., et al. (2016). "Design of a Fire Detection and Control System using Fuzzy Logic."
FireGuard Ecosystem
FireGuard Ecosystem
FireGuard Website
FireGuard Website
FireGuard Website
Unifies live maps, cameras, and sensor data into a single dashboard, allowing residents to cross-reference sources and validate threats instantly.
Unifies live maps, cameras, and sensor data into a single dashboard, allowing residents to cross-reference sources and validate threats instantly.
FireGuard App
FireGuard App
FireGuard App
Delivers real-time, location-based alerts and connects neighbors, turning individual sensor data into a coordinated community response network
Delivers real-time, location-based alerts and connects neighbors, turning individual sensor data into a coordinated community response network
Social Entrepreneurship
Social Entrepreneurship
Empowering Communities, Sustaining Safety
Empowering Communities, Sustaining Safety
Empowering Communities, Sustaining Safety
FireGuard’s mission is to empower every resident with free, verifiable wildfire intelligence, sustained by a community-driven Business Model.
FireGuard’s mission is to empower every resident with free, verifiable wildfire intelligence, sustained by a community-driven Business Model.
FireGuard’s mission is to empower every resident with free, verifiable wildfire intelligence, sustained by a community-driven Business Model.
