ResearvaDirect

• 

  Event Input

• 

  Event Output after Reservation Confirmed

• 

  Login Page

• 

  Page 2

• 

  Page 1

• 

  Landing Page

• 

  Page 3

Inspiration

Booking a table still often means downloading apps, browsing marketplaces, or making phone calls. Many great local restaurants are not listed on reservation platforms at all. We wanted to build a system that could make reservations anywhere in the world without requiring restaurants to adopt new technology. The idea was simple: what if your calendar could act as your personal concierge and make calls on your behalf?

What it does

ReservaDirect transforms Google Calendar into an automated restaurant reservation assistant. Users create a calendar event with preferences like location, cuisine, party size, and price range. The system scrapes nearby restaurant data using Google Maps APIs and leverages Databricks to identify the best possible restaurant based on the given preferences.

If a reservation cannot be confirmed at the top ranked restaurant, the system automatically moves to the next best option, retrying up to five times (configurable). Once a suitable restaurant is selected, the reservation details are sent to ElevenLabs, which places an AI powered voice call to confirm the booking. Any updates or cancellations made in the calendar automatically trigger follow up calls to the restaurant and update the user's calendar accordingly.

How we built it

Frontend: It is hosted on https://calendar-ai-reservations.lovable.app/ (only to be used for the design demo right now, as only our personal google calendars are connected due to security reasons to access the backend) Backend: We integrated Google Calendar webhooks to detect reservation events in real time. Anthropic Claude API based NLP parsing extracts booking preferences from event descriptions. Nearby restaurants are fetched using Google Maps APIs and ranked using a Databricks scoring layer. The selected restaurant is then contacted through ElevenLabs voice agents to confirm reservations. MongoDB stores booking states and user information, while a background workflow handles call scheduling, updates, and retries.

Challenges we ran into

Handling duplicate Google webhook notifications and syncing event updates reliably was complex. Ensuring calls were not triggered during intermediate calendar edits required implementing booking stabilization logic. Managing async voice call workflows alongside real time calendar sync also posed architectural challenges.

Accomplishments that we're proud of

We successfully built an end to end reservation workflow that works without requiring restaurants to join any platform. The system can make, update, and cancel reservations globally using the calendar available phone numbers.

What we learned

We learned how to build event driven systems around third party webhook ecosystems, manage async workflows for real world automation, and design ML ready architecture for decision making layers like restaurant ranking, able to call AI based restaurant calls using ElevenLabs which was highly interesting experience. So, we were able to complete an end to end pipeline for Restaurant booking system. One thing I learnt in my first hackathon was team collaboration and how to deliver quality product in limited time.

Built With

• claudeapi
• databricks
• elevenlabs
• fastapi
• lovable
• mongodb