Google Gemini hackathon entry

₹40 of every ₹100.
That's what an Indian shopkeeper
loses to bad prices — every day.

MARGINS is a phone-based Gemini agent that reads any product, cross-checks GS1 + Agmarknet + ONDC Beckn, haggles in your dialect, and exposes itself as an MCP server any other AI agent in India can call.

Try the live demo →margins-mcp
The problem

India's 63 million small retailers have no oracle for what anything should cost.

12%
average supplier mark-up on FMCG (industry estimate)
90%
of Indian retail still happens at the kirana — but digitally un-served
₹40 of ₹100
lost to bad prices per transaction (conservative estimate)
The solution

MARGINS: 4 steps, 90 seconds, ₹0 infra.

01
Scan

Point the camera at any product. GTIN recognised. Fair price returned in under 4 seconds.

02
Haggle

The supplier overcharges. MARGINS whispers back in your dialect. Settle or walk.

03
Order

Real ONDC Beckn round-trip: search → select → init → confirm. Order id, ACK, the whole JSON-LD.

04
Expose

margins-mcp. Any other AI agent in India can call it as a tool. Infrastructure, not a chatbot.

Under the hood
AI
  • → Gemini 3.7 flash (reasoning + JSON schema)
  • → Gemini 2.5 flash preview TTS (voice)
  • → Gemini Embedding 2 (ledger)
  • → Gemini 3.1 flash lite (routing)
Commerce
  • → ONDC Beckn JSON-LD (4-step round-trip)
  • → GS1 India GTIN lookup
  • → Agmarknet mandi prices
  • → Bhashini (planned)
Infra
  • → Next.js 14 (App Router)
  • → Firebase Firestore (1 GB free)
  • → MCP over HTTP (JSON-RPC 2.0)
  • → Free tier throughout · $0/mo
Try it

One URL. Three demos. ~90 seconds total.

01
Camera + fair price
Real product, real Gemini reasoning, Beckn order button.
02
Haggling scene
Generated script + Tamil/Hindi TTS, settle or walk.
03 ★
margins-mcp
The oracle as infrastructure. JSON-RPC 2.0 over HTTP.
Open the full demo →