CIVII: An AI-Assisted Civilian Initiative

Platform for Post-Disaster and Civic Issue

Reporting
Karthick R Karolin Felix S Yogitha. R
Department of Computer Science and Department of Computer Science and Department of Computer Science and
Engineering, Engineering, Engineering,
Sathyabama Institute of Science and Sathyabama Institute of Science and Sathyabama Institute of Science and
Technology, Chennai, India. Technology, Chennai, India. Technology, Chennai, India.
Karthickkrishna2005@[Link] karolinfelix1704@[Link] [Link]@[Link]

G. Kalaiarasi M. Selvi
Department of Computer Science and Department of Computer Science and
Engineering, Engineering,
Sathyabama Institute of Science and Sathyabama Institute of Science and
Technology, Chennai, India. Technology, Chennai, India.
[Link]@[Link] [Link]@[Link]

Abstract— Civii is a citizen-powered mobile-based climatic variations. Flooding, road collapses, garbage
disaster reporting and rescue support system that overflow, fallen trees, drainage blockages, water leakage,
addresses the communication and coordination gap and electrical hazards are among the most commonly
during emergencies. The application enables users to reported problems in both rural and urban regions. These
capture and report critical incidents using images or incidents significantly disrupt daily life, create threats to
videos directly from their smartphones, with full public safety, and require swift intervention from
support for offline functionality. Reports are stored municipal authorities. Despite the urgency of such
locally when internet access is unavailable and situations, existing reporting mechanisms—including
automatically synchronized with the backend server municipal helplines, field visits, manual complaint
once connectivity is restored. Civii integrates an AI/ML counters, and mobile-based service portals—often suffer
module built with Python (using TensorFlow and from slow processing, lack of verification, and limited
OpenCV) to analyze uploaded media and detect accessibility during emergencies.
emergency conditions such as fire, flood, smoke, or the
presence of people in distress. Based on the AI analysis, The increasing penetration of smartphones and mobile
the system generates and delivers automated rescue internet has created new opportunities for real-time
guidance messages to the user, providing immediate reporting through user-generated multimedia content.
actionable support. Civii also includes a real-time push Civilians today are capable of capturing high-quality
notification system that alerts users within a 1 km radius images and videos that accurately depict local situations.
of any new incident, activating local citizen responders However, current digital platforms rely heavily on manual
and volunteers. In addition to disaster scenarios, Civii review, do not support offline operation during network
supports civic reporting for issues like damaged roads, outages, and lack automated verification techniques. As a
power failures, water leaks, and garbage overflow, result, false reports, irrelevant content, privacy-violating
making it a multipurpose citizen assistance platform. A media, and misuse of public platforms continue to create
secure web dashboard is available for authorized challenges for authorities. These limitations reduce the
responders and administrators to access geo-tagged credibility of crowd-sourced data and delay timely
reports, assess severity, and coordinate appropriate decision-making in high-risk situations
responses. Civii combines AI-driven intelligence, offline
capability, and community collaboration to build a .To address these challenges, this research proposes CIVII
smarter, more resilient emergency and civic support (Civilian Initiative), an AI-assisted mobile and web-based
infrastructure. platform designed to support structured reporting of post-
disaster incidents and everyday civic issues. The system
enables civilians to upload geotagged images or videos,
Keywords— Civic issue reporting, disaster management which undergo two levels of verification. First, an artificial
systems, edge artificial intelligence, geofencing intelligence module performs category validation and
notifications, mobile computing, multimedia verification, unsafe-content filtering through lightweight multimedia
public safety informatics, smart urban services analysis. Second, a human administrator reviews the
verified submissions through a central dashboard to ensure
contextual accuracy before public dissemination. After
I. INTRODUCTION approval, the system generates geofenced push
notifications within a one-kilometer radius to inform
India continues to face recurring natural disasters and nearby app users and automatically dispatches email alerts
widespread civic infrastructure issues as a result of rapid to the corresponding municipal departments.
urbanization, dense population growth, and extreme
The proposed solution emphasizes an offline-first filtering, which are essential to preventing misinformation
architecture, ensuring that users can capture and queue during crises.
reports even during network interruptions caused by
disasters. Once connectivity is restored, the data are Smart city research further extends into improving urban
automatically synchronized with the backend system. This resilience and public safety. Costa [4] analyzed challenges
approach enhances inclusiveness and ensures that critical associated with integrating technology across diverse
reports are not lost due to temporary disruptions. The population groups and identified issues of accessibility,
platform further aims to act as a bridge between civilians digital literacy, and uneven information flow.
and authorities, offering a scalable ecosystem for rapid,
validated, and community-driven reporting. Meanwhile, Bittencourt et al. [5] presented a
comprehensive survey of adaptive smart urban systems
The remainder of the paper is structured as follows. Section that utilize GIS, multi-sensor networks, and cloud-edge
II presents relevant literature in disaster management, civic collaboration. While these studies highlight the importance
reporting, and AI-driven verification. Section III explains of inclusive and adaptive systems, they lack detailed
the proposed system and its core components. Section IV frameworks for incorporating civilian participation with
describes the system architecture. Section V outlines the AI-driven media validation.
methodology and planned algorithms. Section VI discusses
implementation details. Section VII presents expected Despite advancements in disaster prediction, emergency
outcomes based on prior research. Section VIII concludes communication, and smart-city monitoring, existing
the paper and highlights future enhancements. research shows that very few systems combine AI-based
multimedia verification, offline-capable mobile reporting,
II. RELATED WORKS human-in-the-loop admin approval, and geofenced public
Research in disaster management, public safety notification within a single integrated framework. Current
informatics, and civic reporting has expanded rapidly over systems either focus on prediction, communication
the last decade as cities adopt smart technologies to networks, or municipal reporting portals without applying
improve emergency responsiveness. Existing studies have AI to ensure the accuracy and safety of user-submitted
explored diverse approaches, ranging from predictive [Link] analysis of prior research reveals a clear need
modeling and lightweight classification to communication for a platform that supports real-time, validated, and
frameworks intended to enhance situational awareness community-driven reporting. The proposed system, CIVII,
during crisis events. Although these efforts highlight addresses these gaps by integrating lightweight AI models,
significant technological progress, they reveal notable gaps geolocation-based filtering, offline-first synchronization,
in integrating artificial intelligence with community-driven and admin-level moderation to enhance civic issue
reporting for real-time decision support. reporting and post-disaster information sharing.

Several works in the field of disaster prediction rely on

machine learning and hybrid ensemble techniques to III. PROPOSED SYSTEM
analyze environmental patterns. Saleem et al. [1] proposed India continues to experience frequent natural disasters and
the Neural-XGBoost model, a hybrid approach combining persistent civic issues that require timely detection and
neural networks with gradient-boosted trees to predict coordinated response. Despite the availability of municipal
disaster occurrences with improved accuracy. While their helplines, online complaint portals, and dedicated
work demonstrates the potential of hybrid learning emergency services, the current reporting ecosystem
strategies, the system focuses primarily on forecasting and suffers from several critical limitations. First, existing
does not address the challenges of field-level media platforms rely heavily on manual verification, making
reporting or validation. them vulnerable to false complaints, irrelevant
submissions, and inappropriate content, which reduces
Similarly, Raju et al. [2] introduced GeoDisasterAINet, an operational efficiency. Second, many systems do not
ensemble deep learning framework capable of classifying support multimedia-based reporting with automated
various disaster types such as floods, wildfires, and validation, limiting the accuracy and reliability of citizen-
earthquakes using multisource data. Their method utilizes generated data. Third, during and immediately after
convolutional and explainable AI models to support rapid disasters, network failures prevent civilians from accessing
classification. However, the approach operates on online services, resulting in delayed information flow to
structured datasets rather than real-time civilian uploads, authorities. Finally, there is no standardized mechanism for
and therefore does not incorporate user-generated broadcasting verified incidents to nearby individuals who
multimedia or mobile-based reporting workflows. may be affected or capable of providing immediate
[Link], there is a need for a reliable, AI-
Emergency communication systems have also been widely assisted, and community-centered reporting system that
studied. He et al. [3] discussed the use of IoT-assisted validates user-submitted multimedia content, supports
communication infrastructures to support post-disaster offline-first operation, facilitates transparent admin
recovery. Their findings emphasize the importance of moderation, and delivers geofenced notifications to
robust connectivity and buffer-based support in unstable civilians and municipal departments. The problem
network environments. Although such systems ensure addressed in this research is the absence of an integrated
communication continuity, they do not include verification platform that ensures accurate, safe, and rapid reporting of
mechanisms for public submissions or multimedia post-disaster events and everyday civic issues through
intelligent media verification and structured on server-grade and edge-based environments. The AI
communication workflows. module evaluates whether the submitted media matches
the user-selected category and screens content for nudity,
IV. SYSTEM ARCHITECTURE violence, or abusive visuals. The module returns a
The architecture of CIVII is designed as a multi-layer, structured response containing confidence scores and
modular, and scalable system that supports real-time media classification decisions to the backend.
reporting, AI-based verification, administrative approval,
and geofenced notifications. The complete workflow D. Data Storage Layer
integrates a mobile front-end, a cloud-based backend, an
AI verification engine, a centralized administration Firebase Firestore is used to store structured metadata,
dashboard, and communication modules for notification including user information, report details, timestamps,
and email dispatch. The architecture ensures reliability, categories, and verification status. Firebase Storage
extensibility, and support for offline operation, which is maintains the actual media files associated with each
critical in post-disaster scenarios. report. This separation of metadata and media enhances
scalability and reduces query overhead. Additionally,
The overall system architecture is organized into five major Firestore’s real-time update capability enables instant
components: Mobile Application Layer, Backend Service synchronization across the app and admin dashboard.
Layer, AI Verification Layer, Data Storage Layer, and
Admin Dashboard Layer. Each component interacts E. Admin Dashboard Layer
through well-defined APIs to enable seamless data flow
and system coordination. The admin dashboard, implemented using React or Flutter
Web, provides municipal staff or authorized operators with
full visibility into incoming reports. Administrators review
media, verify context, and approve or reject submissions.
Approved reports trigger push notifications and email
alerts, while rejected reports are archived with a
corresponding reason. The dashboard also displays
analytics such as report density, status counts, and
frequently reported categories.

F. Communication and Notification Layer

Fig.1. Architecture Diagram
After an admin approves a report, the system uses Firebase
A. Mobile Application Layer Cloud Messaging (FCM) to send geofenced push
notifications to users located within a one-kilometer radius
The mobile application, developed using Flutter, serves as of the incident. The notification engine uses the Haversine
the primary interface for civilian reporting. Users capture distance formula to compute user proximity based on
images or videos of an incident, select a predefined stored location data. In parallel, Nodemailer dispatches
category, and submit a report along with automatically structured email alerts to the appropriate municipal
captured geolocation metadata. Offline-first capability authorities, including details such as location, category,
enables users to cache reports locally when connectivity is and media links.
unavailable, with automatic upload once the network is
restored. The app communicates with backend services The modular nature of the architecture ensures that each
through RESTful APIs and uses Firebase Authentication subsystem can evolve independently, supporting future
for secure login. enhancements such as severity estimation, multilingual
reporting, and automatic municipal routing. The
B. Backend Service Layer architecture is optimized for reliability in disaster-affected
environments and for scalable deployment across urban
The backend, implemented using [Link] and Express, regions.
manages communication between the frontend, AI engine,
database, and notification systems. The backend receives
user reports, validates request integrity, stores metadata, V. METHODOLOGY
forwards media to the AI module, and updates report status The methodology of the proposed CIVII platform is
based on AI and admin decisions. This layer also handles designed as a structured, multi-stage workflow that ensures
geofencing logic to identify users within a one-kilometer secure, validated, and contextually accurate reporting of
radius of approved reports, and integrates Nodemailer for civic and post-disaster issues. The system incorporates
email dissemination to municipal offices. artificial intelligence, human moderation, geospatial
computation, and cloud-based communication services to
C. AI Verification Layer deliver a complete end-to-end reporting pipeline. This
section outlines the planned methodology, including the
The AI module operates as an independent Python-based data acquisition process, media verification module,
microservice. It performs two core tasks: category administrative approval flow, notification system, and
validation and unsafe-content detection. Lightweight deep offline-first synchronization logic.
learning models such as MobileNet-based classifiers and
TensorFlow Lite models are used to ensure fast inference A. Data Acquisition and User Submission Workflow
The reporting process begins at the user level. Civilians C. AI-Based Category Validation
capture images or videos representing a civic issue or post-
disaster situation using the mobile application. The The AI verification stage ensures that the uploaded media
application collects the following metadata: corresponds to the category selected by the user. A
lightweight computer vision model—such as MobileNet or
Media Content (image/video) EfficientNet-Lite—will be employed to classify the
image/video frame according to predefined civic and
GPS Location Coordinates disaster [Link] AI computes a confidence score for
the predicted label. If the predicted label matches the user-
Selected Category (e.g., flood, road damage, garbage selected category with sufficient confidence, the content is
overflow, lighting failure) marked as Category Verified. Reports with low confidence
scores are flagged for manual scrutiny by the administrator.
User Description (optional text input)
D. Unsafe Content Detection (NSFW Filtering)
Timestamp
To prevent misuse of the platform, a secondary AI model
To ensure usability during network failures, reports are performs NSFW detection. This module identifies:
temporarily stored in a local queue using on-device
caching. Once network connectivity is restored, the queued sexually explicit content
reports automatically synchronize with the backend server.
violent or abusive visuals

privacy-violating recordings

inappropriate or irrelevant media.

Only reports that pass both category validation and NSFW

screening proceed to the administrative review stage. This
two-layer filtering ensures data integrity, improves public
safety, and minimizes the risk of false or harmful
submissions.

E. Administrative Review and Verification

Fig.2. Login A web-based dashboard is provided for municipal staff or

appointed moderators. Each incoming report is displayed
with:
B. Media Upload and Backend Preprocessing
media preview,
After synchronization, the report is transmitted to the
backend server through a secure REST API. The backend AI validation results,
performs two initial tasks:
location on map,
Media Storage in Firebase Storage
user details
Metadata Logging in Firestore (report information,
status, user ID) timestamps.
Once the media is stored, the backend generates a secure The administrator performs the final verification by
URL and forwards it to the AI verification module for marking the report as Approved or Rejected. This hybrid
further processing. AI + human approach balances automation with
accountability.

F. Geofenced Push Notification Dispatch

Upon approval, the system initiates a geofencing algorithm

using the Haversine distance formula. The backend
retrieves all active users within a 1 km radius of the
incident’s location. These users receive a push notification
through Firebase Cloud Messaging (FCM), alerting them
of the nearby [Link] mechanism enhances situational
awareness and encourages community action.
Fig.3. Media Upload
G. Automated Email Alert System
 The backend is implemented using [Link] with Express to
In addition to user notifications, the system sends an provide lightweight, asynchronous REST APIs. These
automated email to the concerned municipal department. APIs handle tasks such as user management, report
The email includes: submission, metadata storage, status updates, and AI
communication. The backend also acts as the gateway
issue description between the mobile app, the AI module, and the database.
It generates email alerts to municipal officers through a
media link dedicated mailing service built using Nodemailer. Error
handling, token-based authentication, and secure media
geolocation access links are incorporated to ensure robustness.

time of detection. C. AI Verification Module (Python)

This ensures that authorities are informed promptly and can The AI component is implemented as an independent
initiate field-level response operations. microservice using Python, TensorFlow Lite, and OpenCV.
Once a report is uploaded, the backend sends the media
H. Issue Resolution Update Workflow link to the AI module, which performs two tasks: (1)
category validation, where a lightweight convolutional
Users or authorities may later update the report status to neural network checks whether the image or video frame
Resolved. This update is reflected across all interfaces and matches the user-selected category; and (2) unsafe-content
archived for future analysis. A resolution timestamp is detection, where a pre-trained NSFW classifier filters
stored for performance tracking. inappropriate or privacy-violating media. The module
returns a verification score and a binary validation flag to
I. Offline-First Synchronization Logic the backend.

Because disasters frequently disrupt communication D. Cloud Infrastructure (Firebase Firestore and Storage)
networks, the system integrates an offline-first behavior:
Firebase Firestore is used to store structured metadata,
Reports are saved locally when offline including user information, report details, timestamps,
coordinates, and admin decisions. Firebase Storage is used
A background service monitors connectivity to securely store uploaded images and videos. The cloud
environment also manages push notifications through
Automatic upload occurs once the network returns Firebase Cloud Messaging. Firestore’s real-time listeners
enable automatic UI updates on both the mobile app and
This increases reliability and ensures no data are lost dashboard.
due to temporary disconnections.
E. Administrative Dashboard (React Web Application)
VI. IMPLEMENTATION DETAILS
The administrative interface is developed using React.
The implementation of CIVII is organized into five major
Administrators can view pending reports, inspect uploaded
subsystems: the mobile application, backend service,
media, verify AI results, approve or reject entries, and
artificial intelligence module, cloud infrastructure, and the change the issue status to “Resolved” when completed. The
administrative dashboard. Although full system
dashboard provides a unified view of all active and past
development is ongoing, each component has been
reports, enabling efficient monitoring and decision-
carefully designed to support modular integration, offline
making. Additional tools support manual notification
functionality, and scalable data processing. broadcasting and email re-sending.
A. Mobile Application (Flutter)

The mobile application is developed using Flutter to ensure

cross-platform support for Android and iOS. The app
enables users to capture images or videos, select an
appropriate issue category, and attach an optional
description. To accommodate disaster scenarios with
unstable connectivity, the application uses a local queueing
mechanism that temporarily stores unsent reports. When
connectivity resumes, the queued reports are automatically
uploaded to the backend. Additional modules include user
authentication, GPS-based location tagging, media
compression, and real-time notification handling using
Firebase Cloud Messaging.
Fig.4. Dashboard
B. Backend Service ([Link] and Express)
 Failure-case testing for offline conditions, weak
F. Integration Workflow networks, and invalid uploads.

The integration process follows a layered approach: the Overall, the system is expected to improve the accuracy,
mobile app uploads the report → the backend stores media safety, and reliability of civilian reporting while enabling
and metadata → the AI module validates the content → the authorities to respond more effectively to post-disaster and
admin dashboard handles review → approved reports civic issues.
trigger notifications and emails. This modular approach
ensures that each subsystem is independently scalable and
easily testable.

VII. EXPECTED RESULT

The CIVII platform is currently under development;
therefore, this section presents the expected outcomes and
planned evaluation procedures based on design objectives,
related literature, and system architecture. The evaluation
focuses on four major dimensions: AI performance, system
efficiency, user experience, and operational reliability
during post-disaster or low-connectivity [Link]
artificial intelligence module is expected to provide
reliable multimedia verification by validating whether an
uploaded image or video corresponds to the selected
category and by identifying inappropriate or unsafe visual
content. Prior studies on lightweight convolutional neural
networks suggest that edge-optimized models can achieve
strong performance for classification tasks. Based on these
findings, the multimedia classifier in CIVII is expected to
obtain a level of precision and recall between 80% and Fig.5. Main UI
90%, depending on the category. The NSFW detection
model is expected to reduce unsafe uploads and maintain VIII. CONCLUSION
platform safety by minimizing false positives and false
negatives during [Link] performance will be The increasing frequency of natural disasters and persistent
evaluated through response time, network handling, and civic infrastructure issues across India highlight the need
notification delivery. The platform is expected to process a for faster, more reliable, and community-centered
complete report—upload, AI validation, and admin review reporting mechanisms. Existing systems remain heavily
request—within a few seconds under stable network dependent on manual communication channels, which
conditions. The geofenced notification system, which uses often fail during emergencies or become ineffective due to
Firebase Cloud Messaging, is expected to deliver alerts to slow processing, limited verification, and poor
users in the defined one-kilometer radius with minimal accessibility. The proposed CIVII (Civilian Initiative)
latency. The offline-first workflow will be assessed by platform addresses these challenges by integrating mobile
measuring its ability to store reports locally and computing, offline-first data handling, and lightweight
synchronize them reliably once network connectivity is artificial intelligence to support authentic and timely
[Link] usability will be evaluated using user reporting of post-disaster and civic [Link] allowing
studies and standard usability scoring methods. The system civilians to upload geotagged images and videos, the
is expected to provide an intuitive interface that simplifies system enables real-time visibility into ground-level
report submission and ensures transparent status tracking. conditions. The incorporation of AI-based category
Admin usability will be examined through dashboard validation and unsafe-content detection further ensures that
responsiveness, media review clarity, and efficiency of the incoming reports are relevant, appropriate, and free from
approval process. misuse. The two-level verification process—AI filtering
followed by admin approval—adds an essential layer of
The evaluation plan includes: trustworthiness before information is shared with nearby
users or municipal departments. The geofenced
Functional testing to ensure correctness of each module; notification system and automated email dispatch
mechanism promote rapid dissemination and encourage
AI performance testing using labeled datasets; faster response actions from authorities.

Load testing on the backend to examine scalability; Although the system is still under development, the
proposed design is expected to improve reporting accuracy,
Notification accuracy testing using geofenced user enhance transparency, and strengthen the collaboration
groups; between civilians and local administration. Future
extensions will include multilingual support, automated
Usability assessments using structured questionnaires; department routing, severity-level estimation, and
expanded AI capabilities for video analysis. Overall, CIVII
has the potential to evolve into a scalable, community-
driven public safety platform that improves civic emergency material demand prediction: Integrating RAG
governance and empowers citizens to actively contribute with reasoning large language models,” IEEE Access, vol.
to disaster resilience and urban development. 13, pp. 167650–167662, 2025.
governance and empowers citizens to actively contribute to
disaster resilience and urban development. [6] Y. He, F. Huang, D. Wang, B. Chen, and R. Zhang,
“Emergency communications in post-disaster scenarios:
IX. REFERENCES IoT-enhanced airship and buffer support,” IEEE Internet of
Things Journal, vol. 12, no. 9, pp. 8783–8798, 2025.
[1] A. S. Raju, S. Sreekanth, R. K. Gatla, M. Rajababu, D. [7] Y. Wang, Q. Hu, Z. Li, Z. Su, R. Li, X. Zou, and J. Zhou,
G. Kumar, A. Flah, C. Z. El-Bayeh, K. A. El-Nagdy, and “Blockchain-envisioned UAV-aided disaster relief
A. Alzaed, “GeoDisasterAINet: An explainable deep networks: Challenges and solutions,” IEEE
ensemble framework for real-time urban and rural disaster Communications Magazine, vol. 63, no. 5, pp. 56–62,
classification and resilience,” IEEE Access, vol. 13, pp. 1– 2025.
18, 2025.
[8] W. Zhou, M. Huang, S. Liu, Q. You, and F. Meng,
[2] D. G. Costa, “Resilient smart cities for everyone: “Research on the construction and application of
Overcoming inequalities in emergency solutions,” IEEE earthquake emergency information knowledge graph
Technology and Society Magazine, vol. 43, no. 4, pp. 45– based on large language models,” IEEE Access, vol. 13,
55, 2024. pp. 145320–145335, 2025.
[3] J. C. N. Bittencourt, D. G. Costa, P. Portugal, and F. [9] M. C. Pognon, A. Quintero, and S. Pierre, “Adaptive
Vasques, “A survey on adaptive smart urban systems,” priority scheduling of Internet of Things data for disaster
IEEE Access, vol. 12, pp. 107120–107138, 2024. management in smart cities,” IEEE Access, vol. 12, pp.
112325–112340, 2024.
[4] M. A. Saleem, A. Javeed, W. Benjapolakul, W. Srisiri,
S. Chaitusaney, and P. Kaewplung, “Neural-XGBoost: A [10] R. Wang, M. L. Chuang, C. Y. Ke, Y. F. Chien, W. J.
hybrid approach for disaster prediction and management Ho, K. C. Chiang, Y. C. Hung, C. H. Tsai, and C. C. Chou,
using machine learning,” IEEE Access, vol. 13, pp. 22510– “Predicting imminent electrical safety incidents using
22525, 2025. smart meter big data with large language models,” IEEE
Access, vol. 12, pp. 102300–102315, 2024.
[5] S. Zhang, M. Huang, S. Liu, F. Meng, Y. Xie, X. Ren,
Y. Zhang, and W. Shao, “AI-driven post-earthquake