Grivence Portal

Inspiration

What it does

How we built it

Challenges we ran into

About the Project Inspiration Public complaints about roads, water supply, electricity, sanitation, drainage, and other civic issues are often manually reviewed and forwarded to the appropriate department. This can lead to incorrect routing, delays, and additional workload for government officials. We wanted to build a system that could automatically understand a citizen's complaint and route it to the right department and officer. This inspired us to develop an AI-based grievance redressal portal that can understand both text and images submitted by citizens. What it does The Grievance Portal allows citizens to submit complaints using text, images, or both. The system:

Analyzes complaint text using a BERT-based NLP classification model.
Processes uploaded images using a YOLO-based image detection model.
Uses OCR to extract useful text from complaint images when required.
Identifies the appropriate government department.
Determines the complaint district from the submitted location.
Automatically assigns the complaint to the appropriate district officer.
Allows officers to view, update, and resolve assigned complaints.
Allows department administrators to monitor complaints and manage officers.
Provides complaint status tracking and notifications. The goal is to reduce manual classification and improve the speed and accuracy of complaint routing. How we built it We developed the system using a modular architecture consisting of a web frontend, backend server, database, and separate AI processing components. Frontend: React.js Backend: Node.js and Express.js Database: MySQL Authentication: JWT Text Classification: BERT Image Detection: YOLO OCR: PaddleOCR Image Processing: OpenCV API Communication: REST APIs The backend handles authentication, complaint management, department and officer management, authorization, and automatic officer assignment. For text complaints, the BERT model analyzes the semantic meaning of the complaint and predicts the appropriate department. For image complaints, the YOLO model detects relevant problems such as potholes, garbage, damaged infrastructure, and other visible civic issues. The AI results are then combined with the complaint location and backend business rules to determine the appropriate officer. Challenges We Ran Into One of our major challenges was collecting and preparing a high-quality dataset for AI training. Public grievance complaints can be expressed in many different ways, making it difficult to create a dataset that represents real citizen language. We also faced challenges in:
Correctly distinguishing between similar departments.
Annotating images for object detection.
Improving image model accuracy.
Handling different image quality, lighting, and viewing angles.
Integrating Python-based AI models with the Node.js backend.
Designing proper authorization for citizens, department administrators, and district officers.
Automatically assigning complaints based on both department and district.
Ensuring that AI predictions could be stored and tracked reliably in the database. Accomplishments That We're Proud Of We successfully developed and integrated the major components of the grievance portal. Our key accomplishments include:
Developed a complete React frontend.
Implemented a Node.js/Express backend.
Designed and implemented the MySQL database.
Implemented JWT-based authentication and role-based authorization.
Successfully integrated frontend, backend, and database.
Developed and integrated the BERT text classification model.
Started and optimized the YOLO-based image processing model.
Implemented automatic department identification.
Implemented district-based officer assignment.
Created separate workflows for citizens, department administrators, and district officers.
Built a complete complaint lifecycle from submission to resolution. What We Learned Through this project, we gained practical experience in both full-stack development and AI/ML integration. We learned how to:
Prepare and clean datasets for machine learning.
Fine-tune a BERT model for text classification.
Prepare and annotate datasets for image detection.
Train and evaluate YOLO models.
Measure model performance using metrics such as accuracy, precision, recall, F1-score, and mAP.
Integrate Python AI services with a Node.js backend.
Design relational databases for real-world applications.
Implement role-based access control.
Separate AI predictions from backend business logic.
Build and test an end-to-end AI-powered application. Most importantly, we learned that building an AI system is not only about choosing a model. Dataset quality, preprocessing, evaluation, integration, and real-world testing are equally important. What's Next for Grievance Portal Our next focus is to improve the accuracy and reliability of the image detection model by expanding and improving the image dataset, refining annotations, and tuning the model. Future improvements include:
Increasing the size and diversity of the image dataset.
Improving YOLO detection accuracy.
Combining text and image predictions for better complaint classification.
Adding confidence-based human verification for uncertain AI predictions.
Improving complaint priority prediction.
Adding multilingual complaint support for regional languages.
Introducing advanced analytics and reporting.
Deploying the complete system for real-world testing.
Continuously improving the AI models using validated complaint data. Our long-term goal is to create a scalable and intelligent public grievance platform that reduces manual effort, improves complaint routing, and helps authorities resolve civic issues more efficiently. ## Accomplishments that we're proud of