CivicLens

RAGHAVENDRA R — Thu, 08 Oct 2026 11:16:37 +0000

CivicLens — Turning Citizen Voices into Actionable Civic Intelligence

Describe the problem. Let AI understand it. Make civic action smarter.

Introduction

Every day, citizens encounter civic problems around them—potholes, garbage accumulation, water leakage, blocked drainage, and broken streetlights. Reporting these problems should be simple, but traditional reporting systems often depend on structured forms and predefined fields.

Citizens don't naturally think in categories such as road_damage, waste, or drainage. They simply describe what they see:

“There is a huge pothole near the bus stand. Bikes are struggling to pass and it has been there for two weeks.”

The challenge is transforming this natural description into information that a digital civic system can understand and act upon.

This is the problem we set out to solve with CivicLens.

💡 What is CivicLens?

CivicLens is an open-source, AI-powered civic issue understanding platform that uses Gemma 4 to transform natural-language citizen complaints into structured civic reports.

Instead of forcing citizens to understand complicated reporting forms, CivicLens allows them to simply describe the problem in their own words.

Our AI analyzes the description and extracts important information such as:

🏷️ Issue category
📝 Issue title
📄 Description
📍 Location
⏱️ Duration
⚠️ Impact
🚨 Priority
🎯 AI confidence

The result is structured information that can serve as the foundation for future civic-management workflows.

🚧 The Problem We Identified

Civic problems are everywhere, but the information surrounding them is often unstructured.

Consider three different citizens reporting the same road problem:

Citizen 1:

“Massive pothole near the bus stand, bikes are finding it difficult to cross.”

Citizen 2:

“Road is damaged badly near the bus stop.”

Citizen 3:

“There has been a big pothole here for almost two weeks.”

All three descriptions communicate useful information, but a traditional system may treat them as completely different text entries.

This creates several challenges:

Unstructured complaints

Citizens describe problems differently, making automated processing difficult.

Lack of standardization

Important details such as category, impact, duration, and priority may not be explicitly provided in a fixed format.

Reporting complexity

Citizens shouldn't need to understand technical categories before reporting a problem.

Difficulty converting reports into actionable information

Raw text is useful to humans, but civic applications need structured data to support future analysis and decision-making.

🤖 Our Solution

CivicLens introduces an AI understanding layer between the citizen and the civic system.

The basic workflow is:

Citizen describes a problem
↓
CivicLens
↓
FastAPI API
↓
Gemma 4
↓
Structured Analysis
↓
Validated Civic Data
↓
User Interface

The citizen only needs to describe the issue naturally.

Gemma 4 handles the understanding and structuring.

🧠 Why Gemma 4?

We wanted AI to be more than just a chatbot in our project.

We use Gemma 4 as a civic issue understanding engine.

Given a citizen's description, the model determines the appropriate civic category and extracts relevant information while following a controlled output structure.

For CivicLens, we currently work with categories including:

road_damage
waste
water
drainage
streetlight
other

The AI also assigns a priority:

low
medium
high

and provides a confidence score between 0 and 1.

This makes the AI output predictable enough for the application to consume programmatically.

🔍 A Real Example

Suppose a citizen enters:

“There is a huge pothole near the bus stand. Bikes are struggling to pass and it has been there for two weeks.”

CivicLens sends the description to Gemma 4.

The resulting analysis can look like:

{
"category": "road_damage",
"title": "Huge pothole near the bus stand",
"description": "A huge pothole is making it difficult for bikes to pass.",
"location": "near the bus stand",
"duration": "2 weeks",
"impact": "Bikes are struggling to pass",
"priority": "high",
"confidence": 0.98
}

Instead of storing only a paragraph of text, CivicLens now has a structured representation of the problem.

⚙️ Technical Implementation

Frontend

Our frontend is built using:

Next.js 16
React 19
TypeScript
Tailwind CSS

The interface is designed around a simple workflow: describe the issue, analyze it, and view the structured result.

Backend

The backend uses:

Python
FastAPI
Pydantic
Uvicorn

FastAPI provides the REST API connecting the frontend with the AI layer.

Our main analysis endpoint is:

POST /api/analyze

The backend validates the incoming civic issue, sends it to Gemma 4, validates the returned structure, and sends the result back to the frontend.

AI Layer

The AI implementation uses:

Gemma 4
Google GenAI SDK
Google AI Studio / Gemini API

We designed the prompting layer to constrain civic categories and priority values while requiring structured JSON output.

🛡️ Validation and Reliability

A major part of the implementation is making sure that AI output can be safely consumed by the application.

We use Pydantic models to validate the response.

For example:

Category must belong to our supported civic categories.
Priority must follow the allowed priority values.
Confidence must remain between 0.0 and 1.0.
Required fields must have the correct structure.

We also validate incoming issue descriptions before sending them to the AI service.

This gives us a clear separation:

Natural Language
↓
AI Understanding
↓
Structured JSON
↓
Schema Validation
↓
Application

🧪 Testing

We tested CivicLens with multiple types of civic complaints.

Examples included:

Citizen Issue AI Category Priority

Pothole near bus stand Road Damage High
Garbage accumulation Waste Medium
Water pipe leakage Water Medium
Blocked drainage Drainage Medium
Broken streetlight Streetlight Medium

This demonstrated that the same AI pipeline can understand different types of civic problems.

🏗️ Architecture

┌──────────────────┐
│ Citizen │
│ Natural Language │
└────────┬─────────┘
│
▼
┌──────────────────┐
│ Next.js UI │
│ React + TS │
└────────┬─────────┘
│
▼
┌──────────────────┐
│ FastAPI │
│ REST API │
└────────┬─────────┘
│
▼
┌──────────────────┐
│ Gemma 4 │
│ AI Understanding │
└────────┬─────────┘
│
▼
┌──────────────────┐
│ Pydantic Schema │
│ Validation │
└────────┬─────────┘
│
▼
┌──────────────────┐
│ Structured Civic │
│ Analysis │
└──────────────────┘

👨‍💻 Team Contributions

Team Member 1 — Problem & Product

Focused on identifying the civic reporting problem, defining the product concept, user workflow, and overall project direction.

Team Member 2 — AI / Gemma 4

Worked on the Gemma 4 integration, civic issue prompts, controlled categories, priorities, structured output, and AI testing.

Team Member 3 — Backend

Worked on the FastAPI backend, API contracts, Pydantic schemas, validation, CORS, and integration between the application and AI layer.

Team Member 4 — Frontend

Worked on the Next.js frontend, civic issue input interface, loading states, analysis result presentation, and frontend-backend integration.

🏆 What We Achieved

During the hackathon, we built a working AI pipeline that can:

Take a natural-language civic complaint → understand it using Gemma 4 → classify it → extract important details → assign priority → return structured civic information.

The most important part of our implementation is that AI is integrated into the actual application workflow rather than being used only as a conversational demonstration.

🚀 What's Next?

CivicLens can be expanded significantly from the current foundation.

🔎 Semantic Duplicate Detection

Multiple citizens may report the same problem using different wording.

Future versions can use semantic similarity to identify potentially duplicate or related reports.

🗺️ Civic Issue Mapping

Structured location information could be visualized on a map to identify areas with repeated civic problems.

👥 Community Clustering

Similar reports could be grouped together to identify larger community-level problems.

📊 Administrative Dashboard

Authorities or administrators could receive a dashboard showing issue categories, priorities, locations, and trends.

🔄 Issue Lifecycle

Future versions could support:

Reported → Under Review → In Progress → Resolved

This would allow citizens to track the progress of reported problems.

🌍 Our Vision

We believe civic technology should not make citizens adapt to machines.

Technology should adapt to the way citizens communicate.

A citizen shouldn't have to know whether a problem belongs to road damage or infrastructure maintenance. They should simply be able to describe what they see.

CivicLens uses AI to understand that description and turn it into structured information that can support future civic action.

🎯 Conclusion

CivicLens is our attempt to bridge the gap between human language and machine-readable civic information.

With Gemma 4, FastAPI, Next.js, and open-source technologies, we created a foundation where a simple citizen description can become structured, categorized, and prioritized civic intelligence.

CivicLens — Describe the problem. Let AI understand it. Make civic action smarter.

🔗 Project

GitHub: github.com/RAGHAVENDRA-11/civiclens-gemma4

Built with: Gemma 4 • FastAPI • Python • Next.js • React • TypeScript • Tailwind CSS • Pydantic • Google GenAI SDK

DEV Community: RAGHAVENDRA R

CivicLens