Second Prize — TechTO Hackathon
Inspiration
Hotel booking looks simple until you try to find the best deal under a deadline. You need to compare prices, verify availability, contact hotels, navigate phone menus, negotiate rates, and secure a reservation.
We built FalconX to automate that workflow using AI voice agents, live pricing data, and a real-time orchestration system.
Our guiding principle was to optimize for three things: speed, price, and distance.
What it does
FalconX takes a natural-language booking request and coordinates the process of finding and booking a hotel.
The system is designed to:
- Parse the request into structured booking details.
- Retrieve live hotel pricing and identify relevant nearby properties.
- Call multiple hotel front desks concurrently using AI voice agents.
- Navigate phone menus, ask about availability, and negotiate competitive rates.
- Compare quotes as they arrive and update the leading option.
- Secure a booking through the available booking workflow and deliver confirmation details.
- Display the process through a live dashboard with call status, transcripts, and price comparisons.
How we built it
1. A backend that orchestrates the workflow
We used FastAPI to coordinate booking requests, pricing, calls, decisions, and notifications.
The request is parsed into structured fields such as destination, check-in and check-out dates, guest count, and email address. These fields drive the rest of the workflow.
2. Voice agents that call real hotels
We used ElevenLabs Conversational AI for outbound voice interactions, with Twilio providing telephony connectivity.
The agents are designed to navigate hotel phone menus, ask for rates and availability, communicate known competing prices, and collect quote and confirmation details.
Structured tool calls allow the backend to process quotes and booking decisions without relying exclusively on free-form conversation transcripts.
3. Parallel calls instead of sequential searching
Calling one hotel at a time creates a bottleneck. FalconX was designed to contact several promising candidates concurrently.
As quotes arrive, the orchestrator updates its view of the best available offer. Failed calls, busy lines, and unavailable rooms can be handled by moving on to other candidates rather than restarting the process.
4. Live pricing and booking integrations
We designed the pricing layer around hotel search data from SerpApi and bookable rates through LiteAPI.
This allows the system to compare online rates with direct hotel quotes. The booking integration also provides a fast path to a reservation while direct calls continue looking for a better deal.
5. A geographically focused hotel database
We prepared a hotel database focused on Ontario, California, and nearby communities. Storing property details and phone numbers in advance reduces the need to discover every candidate during the timed run.
Direct front-desk numbers and mapped phone-menu navigation help reduce delays between identifying a candidate and obtaining a quote.
6. A live interface that shows the work
The web interface was designed to show hotel locations, candidate prices, call progress, live transcripts, and the current leading offer.
Rather than treating the voice agent as a black box, the interface makes the decision process visible as the system moves toward a booking.
How we competed: Speed, price, and distance
Speed: We reduced startup overhead through pre-built hotel data, parallel pricing lookups, and concurrent outbound calls.
Price: We continuously compared incoming quotes against known rates, allowing the system to pursue better offers rather than simply accepting the first response.
Distance: We focused the candidate pool on the target city and surrounding area, keeping the search relevant and avoiding unnecessary work across a much larger geography.
These priorities worked together. A nearby hotel was useful only if it could offer a suitable rate, and a competitive rate mattered only if we could discover and secure it in time.
Challenges we tackled
The main challenge was coordinating several time-sensitive systems: live pricing APIs, outbound calls, voice-agent decisions, booking operations, and confirmation delivery.
We also had to distinguish a conversational response from a reliable booking decision. Structured quote and confirmation data gave the backend a clearer basis for tracking the best option and recording the outcome.
What we're proud of
FalconX demonstrates how voice AI becomes more useful when combined with orchestration, real-time data, and the ability to take action.
The project earned second prize at TechTO Hackathon. More importantly, it gave us a practical opportunity to explore the engineering behind fast, price-aware, geographically focused AI workflows.
Built with
- Backend: Python, FastAPI
- Voice AI: ElevenLabs Conversational AI
- Telephony: Twilio
- Pricing and booking: SerpApi, LiteAPI
- Frontend: Next.js
- Infrastructure: Google Cloud Run
FalconX: Less time searching. Smarter comparisons. Faster booking decisions.
Top comments (0)