Dash2Dorm

Inspiration ✨

During the hackathon, we found ourselves stuck in that all-too-familiar situation—too tired and lazy to leave and grab food, but also hesitant to order from popular delivery apps like Uber Eats because of the high fees and hidden markups. As students ourselves, we knew this was a common problem on campus: food delivery was often expensive, confusing, and didn’t always prioritize fairness for either the customer or the delivery drivers.

This frustration sparked the idea to build Dash2Dorm—a platform designed specifically for campus food delivery that would be transparent, affordable, and flexible. We wanted to create a system with no markups on individual menu items, charging only a small fixed fee and a base courier fee to cover delivery costs. Additionally, by incorporating a dynamic pricing engine that adjusts fees based on real-time factors like weather and demand, we aimed to make the experience fairer and more efficient for everyone involved.

What it does 🚚

Dash2Dorm is a campus-first food delivery platform built to make ordering and delivering food fair, transparent, and efficient for students. Unlike traditional services that inflate menu prices, Dash2Dorm charges only a $0.60 platform fee and a $2 base courier fee, so customers always pay the restaurant’s listed price. Couriers are fellow students delivering on foot, bike, or scooter, making deliveries fast, eco-friendly, and tailored to the campus environment.

Behind the scenes, our dynamic pricing engine uses Gemini on Vellum.ai to predict demand and adjust courier incentives in real time. It analyzes historical delivery data — including timestamps, temperatures, demand levels, and weather — alongside live weather data from Open-Meteo to forecast when and where demand will spike. This lets Dash2Dorm fairly reward student couriers for taking deliveries during tough conditions like rain or high demand while keeping costs low and transparent for customers.

By combining affordability, fairness, and AI-driven intelligence, Dash2Dorm creates a seamless food delivery experience where students can either enjoy meals from campus restaurants without overpaying or earn extra money delivering for their peers.

How we built it 🛠️

We built the full-stack application using Vite+React, TypeScript, Flask, React Native, and utilized multiple APIs like Stripe, Open-Meteo and Supabase to store information, process payments and get real time weather information.

We also added sophisticated route optimization algorithms combining Haversine distance calculations, greedy nearest neighbor with earnings weighting, 2-opt local search heuristics, and geographic clustering. The hybrid approach runs 50 optimization iterations across three strategies to maximize courier earnings per minute while minimizing travel distance. Features multi-objective fitness scoring that balances profitability, time efficiency, and distance penalties. Includes real-time Leaflet map integration with animated route visualization, enhanced markers for pickups/deliveries, and comprehensive route metrics display. Testing shows 10-15% improvement in delivery efficiency compared to basic nearest-neighbor routing, translating to 2-3 additional deliveries per 8-hour shift for increased courier earnings

We built our demand prediction and incentivizing engine using Gemini as our model on Vellum.ai. To get started, we utilized Vellum's AI Workflows to manage our data pipeline, beginning with a realistic mock historical delivery dataset that captured timestamps, temperatures, weather, demand levels, and delivery times. Using Vellum's Document Index, we trained the system to understand the correlation between these variables and make informed predictions . We then connected a real-time Open-Meteo API through a Vellum API Node to get live weather details for the users current location. It fed these details into an AI Prompt Node that queries our indexed dataset to predict demand levels along with expected delivery times. These predicted demand levels are then used as variable across our workflow to power our incentivizing engine Prompt Node which enables dynamic pricing for users and allows for smarter resource allocation.

Challenges we ran into 💥

Our biggest challenge was developing an algorithmic-optimized route for the courier. We decided to use a greedy nearest-neighbour and the haversine distance to calculate the best path. This hybrid approach runs 50 optimization iterations across three strategies to maximize courier earnings per minute while minimizing travel distance. Once that was completed, we also had to draw the route onto the frontend for the user.

Accomplishments that we're proud of

Dynamic Pricing Engine using Vellum: Built a demand prediction model with Gemini on Vellum.ai, combining real-time weather data and historical delivery patterns to adjust courier fees fairly and keep costs low for customers.

Advanced Route Optimization: Developed a hybrid routing system combining Haversine distance, greedy nearest-neighbor with earnings weighting, 2-opt heuristics, and geographic clustering. Running 50 optimization iterations across three strategies, it improves delivery efficiency by 10–15%, enabling 2 to 3 extra deliveries per 8-hour shift and boosting courier earnings.

What we learned

Building Dash2Dorm allowed us to improve our full-stack abilities as a team, and learned how to integrate multiple APIs together. We also learned a lot about Vellum and how useful it can be in AI based projects.

What's next for Dash2Dorm

We want to scale Dash2Dorm to at least 5 different universities across Canada. On top of that, we want to create priority based delivery for couriers that have a higher rating, and work with universities to improve safety protocols. We aim to prepare for formal seed staging by the end of year one.

Built With

• clerk
• gemini
• leaflet.js
• react
• stripe
• supabase
• tailwind
• typescript
• vellum.ai