Conflux

"The flowing together" - where the scattered streams of citizen feedback merge into structured, actionable urban intelligence.

Author: Lothnic

What is Conflux?

Conflux is a civic-tech AI platform that transforms raw, multilingual citizen complaints from social media and public portals into structured, geospatially-aware infrastructure proposals.

The Pipeline:

1. Ingest: Pull civic reports from public, ToS-clean Indian-city sources — Reddit's old public JSON endpoints, civic news RSS feeds, Google News RSS queries for infrastructure complaints, and optionally India's official data.gov.in open-data API. No Reddit OAuth is required.
2. Align: Use multilingual embedding models to map complaints into a shared semantic space.
3. Cluster: HDBSCAN groups complaints by geospatial location and semantic meaning.
4. Analyze: LLM agents generate per-cluster proposals — summary, recommendations, funding sources, responsible agencies, a sequenced communication & outreach plan, an impact/urgency rationale, and an INR budget.
5. Visualize: A Next.js + Leaflet dashboard maps "hotspots" of urban decay, with a deep-dive research engine (satellite context, nearby POIs, policy analysis) that exports a downloadable report.

Tech Stack

• Backend: FastAPI + Uvicorn (code in the app/ package; main.py is a thin entrypoint)
• ML: sentence-transformers, UMAP, HDBSCAN, Groq proposal generation
• Frontend: Next.js 16 (App Router) + Tailwind CSS
• Data: pyproject.toml managed with uv
• Worker: GitHub Actions scheduled ingestion

Quick Start

# 1. Install Python deps
uv sync

# 2. Start the backend
uvicorn main:app --reload --port 8000

# 3. Start the frontend (in a separate terminal)
cd frontend
npm run dev

• Backend API: http://localhost:8000
• Frontend Dashboard: http://localhost:3000

Deployment

See docs/deployment.md for the Render backend and Vercel frontend setup. The backend has a Render blueprint in render.yaml and a lightweight API dependency file in requirements-render.txt.