logo
opening
login
home
identifying
verification
analyzing
pink tax
travel mode
normal price gouging
haggle mode

EquiSpend

Stop getting ripped off. Your AI-Powered Financial Advocate at the Point of Sale.

Inspiration

Consumers face two massive, systemic pricing scams every single day: the Pink Tax (where women's products are priced higher than chemically identical men's versions) and the Tourist Tax (where vendors artificially inflate prices for travelers).

Current financial tools are purely reactive—they just process the swipe after you've already been overcharged. We wanted to build a proactive financial advocate. EquiSpend was created to intercept the transaction before it happens, exposing the real market price and giving consumers the immediate leverage to fight back.

What it does

EquiSpend is a mobile application that detects price gouging in real-time using AI image recognition and live market data.

Pink Tax & Tourist Tax Scanner: Snap a photo of any product on the shelf. The app identifies the item and its gender marketing, then examines live localized market data to find the true fair price or the cheaper male/unisex alternative.
The Verdict: The app instantly shows you the markup percentage, the actual fair price, and a direct buy link to the cheaper alternative so you can act immediately.
Haggle Mode: If you're getting ripped off at a physical market, tap "Haggle This Price." An AI voice agent takes over, speaking directly to the merchant in their native language to negotiate the price down, firmly anchored to the real market data your scan just discovered.

How we built it

We built a complex, full-stack architecture that moves far beyond a simple AI text wrapper:

Frontend: React Native (Expo SDK 54) using TypeScript and Reanimated for a fluid, location-aware native mobile experience.
Intelligence Engine: FastAPI (Python) backend using Google Gemini 2.5 Flash for vision identification and query generation, and SerpAPI to scrape localized Google Shopping data.
Smart Pricing Engine: E-commerce data is incredibly messy and right-skewed. A $7 razor might appear next to a $45 dropshipper bundle. To fix this, we built a custom statistical pipeline:
- Scoring: We actively penalize bulk listings (-10) and gig-delivery apps like Instacart (-5), while heavily boosting first-party retailers like Target (+4).
- Anchor Filter: We drop any price that is > 2.5x the best trusted-retailer price to eliminate scalpers.
- Median vs. Mean: We calculate the final fair price using the median of the filtered dataset, making the app entirely immune to dropshipper outliers.
Haggle Mode Voice Agent: Powered by ElevenLabs Conversational AI, integrated via a React Native WebView. The agent dynamically receives the exact product name, asking price, and our calculated fair market price to ensure it negotiates using hard data, not LLM guesses.

Challenges we ran into

The "Delivery App Anomaly": When pulling market data, local gig-economy apps would frequently return impossibly low prices due to broken-out multi-packs or first-time user coupons. Engineering the negative-scoring algorithm in our pricing engine was critical to finding the true national retail price.
LLM Math Hallucinations in Haggle Mode: Teaching an AI to negotiate mathematically is incredibly difficult. Early iterations of the voice agent would accidentally offer more than the asking price. We had to use strict "Constraint-Based Anchoring" in our prompt engineering to set hard ceilings (Fair Price × 1.15) and dictate how the AI increments its offers.
Cross-Border Complexities: Handling global markets meant dealing with dynamic currency conversions. We integrated the open.er-api to fetch live exchange rates, allowing the app to seamlessly convert 20+ currencies so users can compare foreign tourist prices directly to their home currency.

Accomplishments that we're proud of

Building a complete, highly polished React Native app with a custom Python backend and Supabase caching in under 24 hours.
Engineering a statistical data-cleaning pipeline that actually works on live, unpredictable e-commerce data.
Successfully integrating a real-time, multilingual AI voice pipeline capable of adapting its negotiation style to local cultural norms on the fly.

What we learned

The intricacies of real-world e-commerce data cleaning—specifically why median calculations are vastly superior to mean calculations for pricing models.
Advanced prompt engineering techniques for strict mathematical guardrails in voice agents.
How to orchestrate multiple external APIs (Gemini, SerpAPI, ElevenLabs, Exchange Rates, Supabase) into a single, cohesive sub-second user experience.

What's next for EquiSpend

We want to take EquiSpend from a consumer defense tool to an everyday shopping companion. Our next steps include adding a browser extension for online shopping, expanding our caching system to build a global database of Pink Tax offenders, and refining the Haggle Mode agent to integrate with more native iOS/Android speech-to-text frameworks for offline capabilities.