Smart_economy

RetailIntel — AI-powered location intelligence for Indian retail. Find the best spot to open your store, buy at the lowest price nearby, and see live payment heatmaps by GPS zone.

Comment

Inspiration

India has 63 million small businesses — yet most new shop owners still pick their location by walking around the neighbourhood and asking neighbours. There is no Zillow for retail. No tool that tells a first-generation entrepreneur in Delhi whether Lajpat Nagar or Karol Bagh is the smarter bet for a pharmacy.

We were inspired by a simple observation: every time someone pays for something, they are voting with their wallet on a location. Aggregate enough of those votes onto a map, and you have the most honest demand signal in retail — real money, real places, real time.

What it does

RetailIntel is a three-in-one location intelligence platform:

  1. Store Opener Advisor — an entrepreneur inputs product category, city, and budget. The platform scores every major commercial zone and returns the top 5 locations with an Opportunity Score, risk level, estimated rent, daily revenue range, and a full 12-month profit simulation.

  2. Smart Buyer Finder — a consumer searches for any product. The platform uses their GPS to rank nearby stores cheapest-first, with distance, rating, and open/closed status.

  3. Geotagged Payments — every transaction is stamped with the buyer's GPS coordinates, indexed to an H3 hexagonal cell, and rendered on a live heatmap. Hot zones glow red; emerging zones glow teal. Entrepreneurs can see exactly where purchasing power concentrates before committing capital.

How we built it

Backend — FastAPI + SQLite + H3

The API is a FastAPI application with four routers (/score-location, /simulate, /find-product, /payments) backed by an aiosqlite SQLite database for async I/O.

The Opportunity Score for each zone is computed as:

$$S = \frac{\left(\text{FootTraffic} \times w_t\right) \cdot \left(\text{SpendIndex} \times w_s\right) / 100 \;-\; n_c \cdot \lambda \cdot 5 \;-\; \text{RentIndex} \times 0.4}{1.2}$$

$$\text{Score} = \max(0,\; \min(100,\; S))$$

Where $w_t$ and $w_s$ are category-specific traffic and spend multipliers, $n_c$ is competitor count, and $\lambda$ is the category's competition sensitivity.

Every payment is converted to an H3 hexagonal index at resolution 9 (~174 m cell diameter):

h3_index = h3.geo_to_h3(lat, lng, resolution=9)

The heatmap scales each cell's circle by transaction count:

$$\text{radius}_i = \min(1200,\; 300 + \text{count}_i \times 80) \quad \text{(metres)}$$

$$\text{opacity}_i = \min(0.70,\; 0.15 + \text{count}_i \times 0.06)$$

Frontend — React 18 + Vite + Leaflet

Built with React 18, TailwindCSS v3, react-leaflet for maps, and Recharts for charts. A custom two-layer cursor (8px dot + 36px lagged ring) adds polish. Vite proxies all /api/* calls to the FastAPI backend.

Challenges we ran into

  • H3 API versioning — h3-py 3.x and 4.x have completely different function signatures. geo_to_h3 vs latlng_to_cell caused silent failures until we pinned to h3==3.7.7.
  • Async SQLite on Windows — aiosqlite with the lifespan context manager required careful ordering to ensure DB init completed before the first request.
  • Leaflet + React StrictMode — react-leaflet throws duplicate map container errors under React's double-render. Solved by keeping MapContainer keys stable.
  • Windows drive-letter cd — cd D:\path from a C: prompt doesn't switch drives. Must run D: first, then cd.
  • Empty heatmap on first load — solved with a /payments/seed-demo endpoint that injects 40 realistic geotagged transactions automatically on first mount.

Accomplishments that we're proud of

  • A fully working geotagged payment pipeline — browser GPS → H3 hex index → SQLite → live map circle — built end to end in one session.
  • The scoring formula produces intuitive, defensible rankings: Connaught Place scores high on traffic but is penalised on rent; Rohini wins on cost — exactly what a real entrepreneur would reason.
  • A custom cursor with a lagged ring that expands on hover — purely CSS + requestAnimationFrame, zero libraries.
  • The simulation outputs a breakeven month and a go/no-go recommendation — a concrete decision, not just a score.

What we learned

  • H3 hexagonal indexing is remarkably elegant — fixed-size cells mean no boundary distortion, and resolution 9 gives city-block-level granularity perfect for retail decisions.
  • Browser Geolocation API needs graceful fallbacks — denied permissions are common, so a default coordinate (Delhi centre) is essential.
  • FastAPI lifespan is the correct pattern for DB initialisation — cleaner than the deprecated on_event handlers.
  • Scoring needs category-aware weights — a pharmacy thrives on foot traffic; an electronics store cares more about spend index. One formula fits none.

What's next for Smart_economy

  • Real data ingestion — live feeds from Google Places API, OpenStreetMap Overpass, and government footfall datasets.
  • Heatmap × Score overlay — see in one view whether high-spend zones match high-opportunity zones.
  • WebSocket live stream — push new payments onto the map in real time instead of polling every 15 seconds.
  • User session history — filter the heatmap to your own purchases only.
  • CSV export — download raw payment data from the stats panel.
  • PostGIS migration — swap SQLite for PostgreSQL + PostGIS for true spatial queries and production scaling.

Built With

Share this project:

Updates