nspiration We noticed something at our school. People talk about problems all the time — the broken water fountain, the food waste in the cafeteria, the lack of recycling bins — but nobody actually does anything. It's not because they don't care. It's because nobody knows where to start.
What if you could describe a problem and have AI agents figure out the rest? Not generic advice, but actual root cause analysis, existing solutions from other communities, local people who could help, and a real action plan with a timeline.
What it does You submit a community problem. Five AI agents work through it:
Intake Agent — categorizes, tags, checks urgency Analyst Agent — root causes and systemic issues Researcher Agent — similar cases from other communities Resource Mapper — local organizations, funding, volunteers Solution Architect — builds a full action plan with steps, resources, timeline The agents stream their reasoning live. You watch them think in real-time.
How we built it Backend is Go with SQLite. Each agent is a prompt to the Featherless.ai API running in sequence because each one builds on the last.
Frontend is React 19 + Vite + shadcn/ui + Tailwind CSS v4. We used an open-source dashboard template as the base for the layout and components.
Real-time streaming uses Server-Sent Events from Go to React.
Challenges we ran into SSE was annoying with the Vite dev proxy — EventSource doesn't send custom headers so CORS got weird.
Featherless.ai requires a paid subscription, so we built a mock mode with realistic agent responses. The real API integration is ready to flip on with an API key.
The template we forked had 600+ files deeply tied to Claude Code. Stripping it down took way longer than expected.
Port 8080 was always in use from a previous project. Kept getting bind errors.
Accomplishments we're proud of The agent fleet view — watching five cards populate with live reasoning looks genuinely cool.
The mock solution output reads like a real community action plan. Prompt engineering matters.
Dark glass UI came out cleaner than most hackathon projects.
What we learned Prompt engineering matters more than we thought. A good system prompt is the difference between generic advice and a useful plan.
Go is great for this. Single binary, no runtime deps, SQLite just works.
A working demo with mock data beats a broken demo with real AI every time.
What's next Connect real Featherless.ai API User accounts to track submitted problems Community voting on solutions Mobile version Multi-language support
Log in or sign up for Devpost to join the conversation.