Kolo | Devpost

Story

More than 120 million people worldwide are forcibly displaced by war, disasters, and economic collapse. From Gaza to Ukraine to Sudan and Haiti, communities are rebuilding while global attention moves on.

Fatima is a bakery owner in one of these communities. Despite the crisis, she still opens her shop every morning, but the disaster didn’t just damage buildings. It took away her customers. What Fatima wants is not charity. She wants dignity. She wants a customer.

Nearby live Joseph and Anita, who struggle to access basic necessities like food and clothing. Meanwhile, Diego in Canada sees the news and wants to help, but the only option available is donation. Three people. One problem: no connection between them.

Kolo means “circle.” Diego buys from Fatima, Fatima supports someone in her community, and the story brings the next supporter into the circle.

Most platforms send money. Kolo sends a customer.

Problem Statement

The problem sits at the intersection of three disconnected forces.

Altruistic demand: People want to help communities in crisis. Global charitable giving continues to grow, showing strong willingness to support others. But the only way to support is donation.
Economic collapse for small businesses: When crises occur, local economies suffer first. Small businesses, the backbone of communities, lose customers and revenue.
Community need: At the same time, displaced families struggle to access basic necessities such as food, clothing, and services. Traditional aid systems are often slow, centralized, and disconnected from local businesses that already exist within these communities.

The result is a broken system where people want to help, businesses need customers, and communities need goods, but no platform connects all three.

Different from Competition

Every existing solution sends money. Kolo sends customers.

No platform combines all four:

A named small business you buy from directly
Real revenue - not a loan, not a donation, not a grant
A verified give to someone in that business's community
A proof photo that closes the loop for the buyer

Three Outcomes, One Transaction:

Diego gets proof: a verified impact card, a named person, a real story he can share
Fatima gets a customer: revenue, dignity, and a reason to keep the door open
Joseph and Gabriel get goods: distributed by someone who knows them, not a logistics company

Business model

Revenue Model:

8–10% platform fee on every transaction, taken transparently at checkout
Fee is shown as a line item so, no hidden charges, no surprise deductions
Fee is capped at $15 on any single transaction, regardless of purchase size. Deliberately structured to encourage high-ticket bundles where the business owner captures a greater share

The honest tradeoff:

Yes, small business owner (Fatima) receives less than the full purchase amount
On an 8% fee with a $15 cap: a $25 purchase sends ~$23. A $150 bundle sends ~$135
The alternative is not "Fatima gets 100%". It is "the platform shuts down and Fatima gets nothing"
The fee is the price of the infrastructure that makes the connection possible at all

Tech Overview

Kolo is a Next.js 16 / React 19 application designed to connect buyers with vetted local businesses. It relies heavily on server components and server actions, using Supabase (Auth, Postgres with Row-Level Security, Storage) for the backend and Stripe (Checkout, Identity) for financial and identity operations. The frontend is styled with Tailwind 4 and Framer Motion.

Core Systems & Features:

Proof-of-Fulfillment Engine: To prevent fraud, proof-of-delivery photos go through a rigorous hybrid validation pipeline. It combines deterministic checks (EXIF GPS distance validation against the business location, timestamp checks, SHA-256 duplicate hashing) with an AI vision model that returns a verdict and confidence score.
Dynamic Reputation & Financial Caps: A 90-day rolling engine calculates a 0–100 reputation score based on fulfillment rates, proof validity, buyer ratings, and disputes. This score programmatically dictates a business's monthly revenue processing cap ($0 to $2,000), utilizing a 1.5× growth guardrail to prevent manipulation.
Serverless AI Vetting Pipeline: An asynchronous AWS Lambda function processes browser-recorded onboarding interview. It uses OpenAI Whisper for transcription and GPT-4o-mini to evaluate candidates against a strict six-dimension rubric (e.g., Business Knowledge, Authenticity). This generates a confidence score that heavily influences the admin approval process.
Algorithmic Buyer Feed: The buyer feed uses a sophisticated ranking algorithm that blends trust (35%), recency (25%), engagement (20%), and personalization (15%). A lookahead diversity re-ranker is applied to penalize clustering of the same businesses or categories, ensuring a varied feed.
AI Copywriting & Localized Pricing: Text areas integrate an "Elevate with AI" feature (powered by Qwen3-30B via OpenRouter) to enhance copy while preserving the user's cultural voice and idioms. Additionally, buyer prices are dynamically localized using a blend of Purchasing Power Parity (PPP) and the Big Mac Index for specific regions, making USD amounts tangible locally.
Buyer Impact Dashboard (Kolo Wrapped): After each completed order, buyers can open an animated yearly impact summary showing families helped, businesses funded, countries reached, total value delivered in local purchasing-power-adjusted terms, fastest fulfillment time, top supported business, and a category breakdown (coats, food bundles, school kits). A buyer_wrapped_rank_yearly view computes a supporter percentile across all buyers for the year. Impact rows written at order completion feed the country and category breakdowns directly.
Globe Visualization: An interactive 3D globe (Three.js via a dynamic import) renders all vetted businesses as location pins. Coordinates are sourced from stored lat/lng on the business record, with a Mapbox geocoding fallback for businesses that only have a city/country string. Clicking a pin surfaces the business story, cheapest menu item price, and a direct give link.
Price Anomaly Detection: When a business creates a menu item, the server fetches all existing items across the platform and finds those sharing at least one tag. If the new item's price exceeds 3× the median price of comparable items, it is flagged (price_flagged = true) and a severe_flags_90d reputation penalty is applied immediately. This deters price gouging while allowing legitimate premium listings to exist with reduced visibility.
Live Submitter Geolocation: When a business owner opens the proof-of-completion form, the browser's navigator.geolocation API is called immediately, capturing their current coordinates and reverse-geocoding them to a human-readable city/country via the Nominatim API. This live location is sent to the server alongside the photo as an independent fraud signal, distinct from EXIF GPS embedded in the image. The AI vision model receives both signals and the Haversine distance between submitter and business; the server additionally hard-fails any submission where the submitter is more than 2,000 km from the claimed business location, regardless of the AI verdict.