FalconX: How AI Voice Agents Are Reshaping Hotel Bidding and Booking

Starting with "calling around for prices": the automation gap in hotel booking

Hotel booking looks straightforward—search, compare, contact, confirm—but executing it is full of friction. You switch between OTA platforms to compare prices, call the front desk to confirm availability and negotiated rates, navigate through phone menus layer by layer, and still have to close the deal within a limited window. The TechTO Hackathon second-place project FalconX targets exactly this gap: it replaces manual phone calls with AI voice agents, handing the entire chain of "search—compare—negotiate—book" to a real-time orchestration system.

According to the team's explanation published in the rss source, FalconX's design principles boil down to three measurable goals: speed, price, distance. These three words determined every subsequent technical choice—from the pre-built hotel database to the parallel dialing architecture, from structured quote parsing to the geofencing strategy.

Architecture breakdown: the orchestration layer is the real "brain"

The most noteworthy thing about FalconX isn't which voice model it uses, but that it treats the voice agent as an orchestratable component rather than a standalone conversational bot.

Separation of backend orchestration and voice channel

The system uses FastAPI as its orchestration core, parsing natural-language booking requests into structured fields—destination, check-in/check-out dates, number of guests, email address. These fields drive all subsequent actions: retrieving hotels, initiating calls, comparing quotes, executing bookings. The voice interaction layer is handled by ElevenLabs Conversational AI, with Twilio providing the telephony connection.

The key to this layered design is that while the voice agent handles "conversation," the processing of quotes and confirmations relies on structured tool calls rather than pure text transcription. The original project write-up specifically emphasizes that this lets the backend distinguish between "conversational responses" and "reliable booking decisions"—in a time-sensitive bidding scenario, this distinction is the prerequisite for whether the process can advance automatically.

ElevenLabs Conversational AI is a voice agent platform launched by ElevenLabs, built on its text-to-speech (TTS) and speech recognition (ASR) capabilities, allowing developers to build real-time conversational voice bots via API. Compared to traditional IVR (interactive voice response) systems, it supports more natural turn-taking, interruption handling, and context retention. In FalconX, it handles voice interaction with hotel front desks, but the key business logic (such as quote parsing and booking decisions) does not depend on its conversational output—instead, it relies on structured tool calls made by the backend orchestrator. This division lets the voice layer focus on conversational experience while the business layer ensures decision reliability and traceability.

Twilio is a cloud communications platform offering programmable phone, SMS, and video APIs. In this project, Twilio mainly provides outbound telephony lines, enabling FalconX to programmatically initiate and manage multiple parallel phone calls. Through Twilio's API, the orchestrator can dynamically control call initiation, hang-up, and status monitoring—the infrastructure underlying the parallel dialing strategy.

Parallel dialing replaces serial search

Calling hotels one by one creates an obvious bottleneck. FalconX's strategy is to contact multiple high-potential candidate hotels simultaneously; as quotes return, the orchestrator continuously updates the current best offer. Busy lines, failed calls, and no-availability responses don't require restarting the process—just move on to the next candidate. This design converts the "wait—retry" loss common in phone scenarios into parallel probabilistic coverage.

Dual-track pricing: online data and direct phone sourcing

The pricing layer draws from two sources at once: SerpApi provides hotel search data, and LiteAPI provides bookable rates. Online data and front-desk phone quotes cross-reference each other—the system can use known online prices as leverage to negotiate competitive rates with the front desk; the booking integration also provides a fast path to lock in a reservation, while continuing to seek better quotes by phone.

SerpApi is a real-time search API that scrapes structured results from search engines like Google. In the hotel scenario, it can aggregate hotel listings, prices, and availability from multiple OTA (online travel agency) platforms, providing FalconX with an initial set of candidate hotels and online reference prices. LiteAPI, meanwhile, is a hotel booking API that provides directly bookable inventory and rates and supports completing booking operations. The combination forms a "search discovery + booking execution" loop: SerpApi helps the system discover candidates and obtain publicly available online prices, while LiteAPI provides a fast path to lock in reservations and its price data also serves as a comparison benchmark when negotiating with front desks. This dual-track design strengthens the system's information advantage in phone negotiations.

Geographic focus and real-time visualization: two underrated design decisions

A pre-built hotel database narrows the search space

FalconX prepared a hotel database focused on Ontario, California and surrounding communities, pre-storing hotel details and phone numbers. This decision directly serves the "distance" goal: within a limited time, the system doesn't need to discover every candidate hotel on the fly, and it reduces the latency between identifying candidates and obtaining quotes. Front-desk direct dial numbers and pre-mapped phone menu navigation further compress operational time.

A detail worth noting: geographic focus isn't just an efficiency optimization—it also makes "distance" a variable that can participate in bid ranking. Nearby hotels only matter if they offer a suitable price, and a good price is only valuable if it can be discovered and locked in within the time limit.

Making the voice agent no longer a black box

The web interface (built with Next.js) displays hotel locations, candidate prices, call progress, call transcripts, and the current leading offer in real time. In voice AI applications, this design deserves separate discussion: voice interaction naturally lacks observability—users can't understand "what the system is doing" the way they watch a webpage loading bar. FalconX uses a real-time dashboard to externalize the decision process, letting users see how each dial, each conversation, and each quote moves the system closer to the final booking.

Engineering challenges and reusable lessons

The team noted that the biggest challenge was coordinating multiple strongly time-sequenced systems: real-time pricing APIs, outbound calls, voice agent decisions, booking operations, and confirmation delivery. Any delay or failure in any link could invalidate the entire bidding window. Their approach was to base key decisions on structured data rather than relying on semantic understanding of free-form conversation.

This lesson has direct reference value for teams building similar voice agent applications: voice models handle the interaction experience, but the business loop should be driven by verifiable structured events. When the system needs to choose among multiple competing quotes, traceable data matters more than fluent conversation.

Tech stack overview

• Backend: Python, FastAPI
• Voice AI: ElevenLabs Conversational AI
• Telephony: Twilio
• Pricing and booking: SerpApi, LiteAPI
• Frontend: Next.js
• Infrastructure: Google Cloud Run

FalconX won second place at the TechTO Hackathon. The path it demonstrates is clear: when voice AI is placed inside a system with real-time data, orchestration capabilities, and executable actions, it truly transforms from "can converse" to "can get things done." For scenarios like hotel booking—where information is scattered and decision deadlines are clear—an architecture of parallel dialing plus structured quotes may offer more practical value than simply optimizing the conversational naturalness of the voice model.

One-line Summary

FalconX uses AI voice agents to call hotels in parallel, automating the entire search, price comparison, negotiation, and booking process.

Paragraph Summary

FalconX is a second-place project at the TechTO Hackathon aimed at replacing manual hotel calls with AI voice agents to automate search, price comparison, negotiation, and booking. The system uses FastAPI as its orchestration core, parsing booking requests into structured fields that drive hotel retrieval, parallel dialing, quote comparison, and booking execution. The voice interaction layer is handled by ElevenLabs Conversational AI, with Twilio providing telephony. Key design choices include parallel dialing instead of serial search, dual-track pricing via SerpApi and LiteAPI, a pre-built hotel database for the Ontario, California area to reduce latency, and a real-time visualization dashboard showing the decision process. The project emphasizes structured tool calls rather than free-text transcription to ensure reliable booking decisions.