CJP – Civic Journey Platform
Introduction
What happens after a citizen reports a civic issue?
In many cases, complaints are submitted, acknowledged, and eventually forgotten. Citizens often find themselves reporting the same problem repeatedly because there is no persistent memory connecting related reports, actions, and outcomes.
To address this challenge, I built CJP (Civic Journey Platform), an agentic civic intelligence platform that transforms citizen reports into persistent civic issues with accountability tracking, semantic search, and actionable resolutions.
Vision and What It Does
The goal of CJP is to ensure that citizen concerns do not disappear after submission.
Instead of treating every report as an isolated event, the platform creates a persistent civic memory where related reports, evidence, actions, and outcomes remain connected over time.
Key capabilities include:
- Natural language civic issue reporting
- AI-powered issue understanding
- Semantic search across related issues
- Persistent memory across sessions
- Accountability and action tracking
- Practical recommendations and job-matching workflows for employment-related concerns
The platform focuses on helping citizens move from reporting problems to tracking meaningful progress.
How I Built It
I started by designing the platform around an agent-first architecture rather than a traditional chatbot experience.
The frontend was developed using React, TypeScript, Tailwind CSS, and Vite, while FastAPI powers the backend APIs.
For intelligence and reasoning, I used the Strands Agents SDK together with Amazon Bedrock. The agent dynamically selects tools based on user intent, enabling more flexible and contextual interactions.
One of the biggest challenges was implementing persistent memory. I wanted users to be able to return later and continue discussions without losing context.
To achieve this, I integrated CockroachDB Cloud as the memory layer for storing civic issues, evidence, timelines, actions, and semantic embeddings.
Another challenge was connecting similar reports that use completely different wording. This led me to implement vector-based semantic search using embeddings generated through Amazon Titan Embed V2.
AWS Services Used
Amazon Bedrock
Amazon Bedrock serves as the intelligence layer of the platform.
Services used:
- Amazon Nova Pro for reasoning and agent interactions
- Amazon Titan Embed V2 for vector embeddings and semantic search
These services enable the platform to understand citizen concerns, retrieve relevant context, and provide intelligent responses.
Architecture Overview
Citizen → React Frontend → FastAPI Backend → Strands Agent
The agent communicates with Amazon Bedrock for reasoning and embeddings while leveraging CockroachDB Cloud as a persistent memory layer.
This architecture enables:
- Cross-session memory retrieval
- Scalable AI-powered workflows
What I Learned Across the Summer
Building CJP taught me several valuable lessons:
AI Agents Need Memory
Persistent memory transforms an assistant from a simple chatbot into a system capable of long-term reasoning and context retention.
Semantic Search Improves User Experience
Embeddings allow applications to understand meaning rather than rely on exact keyword matching.
AWS Accelerates Innovation
Amazon Bedrock significantly reduced the complexity of integrating foundation models and embeddings, allowing me to focus on solving real-world problems.
Accountability Matters
Technology becomes more impactful when it helps users track progress and outcomes rather than simply record information.
Future Improvements
I plan to continue expanding CJP with:
- Government department integrations
- Enhanced analytics dashboards
- Advanced evidence verification
- Broader civic issue categories
Project Links
Live Application
GitHub Repository
Loading article