Delivry

1. Inspiration

Ordering food—especially with allergies, big groups, or vague cravings—is slow, annoying, and full of trial-and-error. We wanted a system that could understand natural language like “quesabirrias, burgers, and horchata for 12 people,” handle allergies safely, and instantly assemble the perfect order. So we built Delivry.

1. What it does

Delivry is an AI food-ordering agent that turns any request into a ready-to-checkout meal plan. It can:

Parse cravings, budgets, locations, allergies, and dietary rules

Search real nearby restaurants

Strictly filter out unsafe or allergen-containing spots

Build single recommendations or multi-restaurant batch orders for events

Provide direct checkout links

Validate safety using Daytona sandbox runs

1. How we built it

We combined:

OpenAI LLMs to parse user intent and pick the best restaurants + items

Google Places API for real restaurant data

Strict allergen + component filtering (dairy → milk, cheese, butter, etc.)

Daytona to run Python snippet validations inside secure sandboxes

FastAPI backend + lightweight HTML/CSS/JS frontend

A multi-restaurant “batch order” pipeline designed and tuned during the hackathon

1. Challenges we ran into

Getting LLMs to truly respect allergies—especially hidden components

Preventing “hallucinated” menu items

Handling vague user prompts without returning empty results

Merging multi-restaurant logic with strict safety rules

Hitting Daytona sandbox limits and restructuring our validation flow

Keeping the frontend responsive while adding batch-order features

1. Accomplishments that we're proud of

Fully working AI-to-checkout ordering system

Robust allergen-aware filtering that actually works

Automatic batch ordering across multiple restaurants

Integrating Daytona, Galileo, for validation + debugging

Clean UI that makes the whole experience look effortless

1. What we learned

LLMs need very explicit constraints for safety (especially allergies)

Daytona can act as a powerful real-time validator for AI decisions

Group order planning is surprisingly hard without specialized logic

Frontend/Backend iteration speed is everything in a hackathon

The best UX is one where the user can type anything and get instant value

1. What's next for Delivry

Multi-checkout auto-completion for DoorDash/UberEats/Grubhub

True menu scraping + price estimation

Profiles with saved allergies, preferences, and favorite cuisines

Automatic “refresh suggestion” button powered by LLM diversification

Delivery-time optimization across multiple restaurant orders

Turning Delivry into a full standalone AI food-ordering platform

Built With

• api
• daytona
• intel-galileo
• java
• javascript
• python
• sentry
• vscode