Inside Civic Sathi: An AI Pipeline for Municipal Grievances
How one pothole report travels from a citizen's photo to a sealed-bid work order — vision triage, deduplication, a database-aware copilot and cross-department conflict detection.
- Published
- Nov 2026
- Reading
- 7 min
- Topics
- civic tech, computer vision, LLM, SIH 2026
The problem
Indian cities don't lack complaints — they lack triage. Reports arrive through scattered channels, get routed by hand, duplicate each other, and disappear into departments without a visible trail. Civic Sathi, built by Team Vertex for Smart India Hackathon 2026, treats the grievance as a pipeline instead of a ticket.
1 · The citizen report
A citizen uploads a photo. A multimodal vision model classifies the issue (for example, road damage), estimates severity and suggests the responsible department — so the citizen never has to understand municipal org charts. Geofencing rejects pins outside the municipal boundary, and every report receives a tracking ID and a public timeline. The portal supports English, Hindi, Gujarati and Kannada.
2 · Municipal triage
Officers work from a GIS ward heatmap with SLA tracking. The most valuable feature is deduplication: dozens of reports about the same pothole are merged into one canonical civic issue, so the queue reflects problems, not noise.
A copilot answers questions such as "how many open complaints are there in this ward?" from a live database snapshot rather than from a model's memory — the difference between a chatbot and a decision tool.
3 · Tendering and proof of work
Verified issues generate tenders. Contractors submit sealed bids that stay closed until the opening time, which removes the incentive to collude. After award, milestones are backed by geo-tagged before-and-after photos that are reviewed before payment.
4 · Cross-department prevention
The admin layer looks across departments for collisions — a road resurfacing scheduled in the same ward as an underground water-pipe repair. Sequencing the water work first avoids digging up a new road weeks after it was laid.
What I took away
AI is most useful here as plumbing: classification, merging and summarising. The trust comes from the boring parts — role-segregated portals, audit trails, sealed bids and photo evidence.