/ projects / sarvik
Every petition,
sorted.
Thousands of petitions reach an MP. Nobody can tell which ones the government already pays for. So the ₹5 crore constituency fund pays twice for the same road, and the quiet villages never surface at all.
01 / The problem
Requests arrive as voice notes, paper letters, WhatsApp forwards and portal submissions, across five languages. There is no way to see that four hundred people asked for the same thing, and no way to check whether a central scheme already funds it. The money gets spent on whoever shouted loudest, and the family still waiting never appears in the data at all.
/ See it work
The full walkthrough. Watch on YouTube ↗
02 / How it works
01
One inbox
Voice notes, photographed letters and text become one structured record. No language barrier.
02
Forward, don't spend
Each petition is matched to the exact department and central scheme, so the fund stays untouched for real gaps.
03
No duplicate works
Identical requests get merged, and infrastructure that already exists nearby gets flagged.
04
Ranked on real costs
Remaining gaps are priced from government data, so the decision runs on numbers rather than volume.
05
The silent are heard
Government data surfaces real needs in quiet villages, even where nobody sent a petition.
06
Answered in their language
Each petition is tracked to the end, and the citizen gets the reply in their own language.
03 / Results
in the People's Priorities track of Build with AI: Code for Communities, by Google and Hack2skill. 7,000+ people took part.
- ₹1 lakhprize for the track win.
- 22 Julshown to MPs at The Imperial, New Delhi, a week after the win.
Measured Jul 2026
04 / How it's built
- The model reads. The data decides.Every figure comes from government data, and every decision stays with the MP. An AI that invents a cost estimate for public spending is worse than no AI.
- Built for any constituencyIt was built on Coimbatore's data. Adding another constituency means loading its data, not rebuilding.
- Need is measured, not just heardPetitions favour whoever asks loudest. So government data is used to find needs in quiet villages that never send one, and every gap is ranked on real cost.
Stack