Skip to content

KrishMaske/EquiSpend

BranchesTags

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

27 Commits
backend
backend
 
 
docs
docs
 
 
frontend
frontend
 
 
.gitignore
.gitignore
 
 
README.md
README.md
 
 

Repository files navigation

EquiSpend - Fiserv Challenge & Google Gemini Challenge Winner @ HackHERS

Stop getting ripped off. Snap a photo of any product, and EquiSpend tells you the real price — instantly.

EquiSpend is a mobile app that detects price gouging in real time using AI image recognition and live market data. It catches the two most common pricing scams consumers face every day:

  • Pink Tax — Women's products priced higher than identical men's versions (same formula, different color, bigger price tag)
  • Tourist Tax — Vendors inflating prices for travelers who don't know what things cost locally

It also includes Haggle Mode: an AI voice agent that negotiates with the merchant for you, in their language, anchored to the real market price.

How It Works

  1. Snap a photo of a product on the shelf
  2. AI identifies the brand, product, category, and gender marketing (Gemini 2.5 Flash)
  3. Live price search pulls real Google Shopping data for your region (SerpAPI)
  4. Pricing engine scores results, filters junk listings, and calculates the true fair price
  5. Verdict — see the markup percentage, fair price, cheaper alternative with a buy link, and a one-tap button to haggle
 📷 Scan ──▶ 🤖 Identify ──▶ 💵 Confirm Price ──▶ 📊 Results ──▶ 🤝 Haggle

Features

Feature What it does
🚺 Pink Tax Detection Compares female-marketed products to male/generic equivalents. Same razor, same gel — exposes the gender markup.
🌍 Tourist Tax Detection Searches the local market in your actual location to find the real price vs. inflated tourist price.
🤝 Haggle Mode AI voice agent negotiates with the shopkeeper in their language. Anchored to the fair market price from your scan. Uses ElevenLabs Conversational AI.
💱 Multi-Currency 20+ currencies. Live exchange rates. Prices auto-convert so you can compare across borders.
📊 Smart Pricing Engine Median-based pricing. Anchor filter (2.5x cap). Penalties for bulk listings and delivery apps. Resistant to outliers.
🔗 Buy Links Every alternative includes a clickable link with thumbnail — act on it immediately.
💾 Caching Results cached in Supabase with currency tags. Faster repeat lookups, fewer API calls.
📜 Scan History Every scan saved locally with timestamps, prices, and location.
📍 Location-Aware GPS auto-detection + manual override. SERP results localized by country code.

Haggle Mode

After scanning, if a product is overpriced, a "Haggle This Price" button appears on the results card. Tap it, and an AI voice agent connects via ElevenLabs Conversational AI to negotiate with the merchant directly.

What makes it work:

  • The agent receives the exact product name, asking price, and fair market price from your scan — it's not guessing
  • Opens at ~70% of fair price and grinds up in tiny increments. Hard ceiling at fair price × 1.15. Walks away if the merchant won't budge
  • Speaks the merchant's language — detects what language they use and mirrors it instantly
  • Adapts negotiation style to local culture (relationship-focused in Latin America, direct in Germany)
  • Runs inside a WebView (ElevenLabs convai widget) — works in Expo Go, no native build required

The full agent configuration (system prompt, client tools, voice settings) is in docs/elevenlabs-agent-config.md.

Pricing Engine

The price comparison isn't a simple average. The pipeline:

Gemini identifies product → generates comparison search query
        ↓
SerpAPI hits Google Shopping (localized by country)
        ↓
Results scored:
  +4  major retailer (Amazon, Walmart, Target)
  +3  brand match
  +2  keyword overlap
  -10 bulk/multipack ("pack of 12")
  -5  delivery app (Instacart, DoorDash)
  ✕   filtered: refurbished, accessories, knockoffs
        ↓
Anchor Filter: drops prices > 2.5× the best trusted-retailer price
        ↓
Median of remaining prices (immune to dropshipper outliers)
        ↓
Currency conversion if needed (live FX rates)
        ↓
Fair price + markup % + savings amount + buy link

Why median, not mean: E-commerce data is right-skewed. A $7 razor appears next to a $45 dropshipper bundle. The mean gets pulled up; the median finds the truth.

Why the anchor filter: If the best trusted-retailer price is $7, anything over $17.50 (2.5×) gets dropped — bulk packs and scalpers eliminated before they inflate the benchmark.

Architecture

┌────────────────────────────────────────────────┐
│  FRONTEND — React Native (Expo SDK 54)         │
│                                                │
│  Scanner → Results → History → Profile         │
│  expo-camera    Reanimated    AsyncStorage     │
│  expo-location  expo-image-picker              │
│                                                │
│  Talk (Haggle Mode)                            │
│  └─ WebView → ElevenLabs convai widget         │
└──────────────────┬─────────────────────────────┘
                   │ REST API
                   ▼
┌────────────────────────────────────────────────┐
│  BACKEND — FastAPI (Python 3.10+)              │
│                                                │
│  POST /scan/identify  →  Gemini 2.5 Flash      │
│  POST /scan/analyze   →  Gemini → SerpAPI      │
│                          → Pricing Engine       │
│                                                │
│  Gemini (Vision + Query Gen)                   │
│  SerpAPI (Google Shopping)                     │
│  open.er-api.com (Live FX Rates)               │
│  Supabase (PostgreSQL — cache + auth)          │
└────────────────────────────────────────────────┘

Tech Stack

Layer Tech
Mobile App React Native 0.81, Expo SDK 54, TypeScript, expo-router
UI / Animation react-native-reanimated, Animated API
Camera / Location expo-camera, expo-image-picker, expo-location
Haggle Mode ElevenLabs Conversational AI (convai widget via react-native-webview)
Backend FastAPI, Python 3.10+, Uvicorn
AI Google Gemini 2.5 Flash (vision + text generation)
Price Data SerpAPI (Google Shopping, localized)
Database Supabase (PostgreSQL) — product cache + user auth
Currency open.er-api.com (live exchange rates)

Project Structure

EquiSpend/
├── backend/
│   ├── app.py                    # FastAPI entry point
│   ├── config/settings.py        # API keys (Supabase, Gemini, SerpAPI)
│   ├── routes/
│   │   ├── scanner_route.py      # /scan/identify + /scan/analyze
│   │   ├── auth_route.py         # Auth endpoints
│   │   └── db_route.py           # Database ops
│   └── utils/
│       ├── gemini.py             # Gemini prompts & analysis
│       ├── logic.py              # Pricing engine (SERP, scoring, cache)
│       ├── prompt.json           # Primary Gemini prompt
│       └── failover_prompt.json  # Fallback prompt
│
├── frontend/
│   ├── app/
│   │   ├── scanner.tsx           # Camera + filter select + location
│   │   ├── results.tsx           # Results cards + Haggle button
│   │   ├── talk.tsx              # Haggle Mode (ElevenLabs WebView)
│   │   ├── history.tsx           # Scan history
│   │   ├── login.tsx / signup.tsx
│   │   └── profile.tsx
│   ├── constants/
│   │   ├── api.ts                # Backend URL
│   │   ├── currencies.ts         # Currency codes + symbols
│   │   └── theme.ts              # Colors
│   └── image-store.ts            # Shared state (image, location, price)
│
├── docs/
│   └── elevenlabs-agent-config.md  # Haggle Mode dashboard config
└── README.md

Quick Start

Prerequisites

Backend

cd backend
python -m venv venv && venv\Scripts\activate  # Windows
# source venv/bin/activate                    # macOS/Linux
pip install -r requirements.txt

Create backend/.env:

GEMINI_API_KEY=your_key
SERP_KEY=your_key
SUPABASE_URL=your_url
SUPABASE_KEY=your_key
uvicorn app:app --reload --host 0.0.0.0 --port 8000

Frontend

cd frontend
npm install
npx expo start

Scan the QR code with Expo Go. Both devices must be on the same network — update frontend/constants/api.ts with your machine's IP, or use ngrok for tunneling.

Haggle Mode Setup

  1. Create an agent at elevenlabs.io/app/conversational-ai
  2. Paste the system prompt and client tool config from docs/elevenlabs-agent-config.md
  3. Set the Agent ID in frontend/app/talk.tsx

API

POST /scan/identify

Send a product photo, get back the identified product.

{
  "brand": "Gillette Venus",
  "product_name": "Smoothing Cleanser + Shave Gel",
  "category": "Shave Gel",
  "volume": "6.42 oz",
  "gender_marketing": "women"
}

POST /scan/analyze

Run the gouging analysis. Pass mode (girl, travel, general, or both), product info, user price, currency, and location.

{
  "product_name": "Smoothing Cleanser + Shave Gel",
  "price_scanned": 9.99,
  "fair_price": 6.99,
  "equity_gap": 3.00,
  "comparable_product": "Gillette Series Sensitive Shave Gel",
  "source": "Target",
  "suggestion_link": "https://...",
  "suggestion_image": "https://..."
}

Supported Regions

Works anywhere with Google Shopping coverage. Localized via country codes (gl=us, gl=in, gl=fr, etc.). Currency conversion for 20+ currencies via live FX rates.

Region Status
US, Canada, UK, Australia Full support
India, Japan, Germany, France Full support
Any other Supported via FX conversion fallback

Team

Built by Team EquiSpend — because price gouging shouldn't be invisible.

License

Educational and hackathon purposes.

About

No description, website, or topics provided.

Resources

Readme
Activity

Stars

0 stars

Watchers

0 watching

Forks

1 fork
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages