RunHotel
HospitalityAI-Powered Hotel Management with Voice Commands
60% faster check-in with Hindi voice AI
The Challenge
Budget hotels in India run on Excel sheets and outdated PMS software. Staff are non-technical, multilingual (Hindi/English), and need hands-free operation during busy check-in hours. Existing cloud-based voice assistants failed in this environment — high latency over unreliable internet, no support for Hindi/Hinglish code-switching, and privacy concerns with sending guest data to third-party servers.
- Staff literacy barriers made traditional software interfaces ineffective
- Internet connectivity in tier-2 and tier-3 Indian cities is unreliable
- Existing PMS solutions were designed for large hotel chains, not budget properties
- Hindi and Hinglish voice recognition was unavailable in any commercial PMS product
Our Approach
We built a complete AI-first PMS with on-device voice commands. A custom on-device SLM runs locally using ONNX Runtime for zero-latency Hindi/Hinglish processing — no cloud dependency means it works even with poor internet. The Flutter mobile app serves as the primary interface, while a Next.js dashboard provides property owners with analytics and multi-property oversight.
- Trained a compact language model on hotel-specific Hindi/Hinglish voice commands
- Implemented on-device inference via ONNX Runtime — 120ms end-to-end latency
- Built Flutter mobile app for staff with voice-first UX and minimal text input
- Created Next.js admin dashboard for property owners with occupancy analytics
- Designed offline-first architecture with background sync when connectivity returns
Key Features
Hindi Voice Commands
Staff speak naturally in Hindi, Hinglish, or English to check in guests, assign rooms, and manage billing — no typing required.
On-Device AI Processing
Custom SLM runs entirely on the device via ONNX Runtime. Zero cloud latency, works offline, and keeps guest data private.
Multi-Property Dashboard
Property owners monitor occupancy, revenue, and staff activity across multiple hotels from a single Next.js web dashboard.
Offline-First Architecture
All core operations work without internet. Data syncs automatically when connectivity returns — critical for tier-2/3 cities.
Key Results
120ms
Voice Latency
End-to-end from speech to action
500+
Commands/Day
Processed daily across pilot hotels
3
Languages
Hindi, Hinglish, English
60%
Check-in Speed
Faster than manual process
“The voice commands changed everything. Staff who couldn't type can now manage the entire hotel by speaking.”
Hotel Manager
Pilot Hotel, Delhi NCR
Frequently Asked Questions
How does the voice AI work without internet?
RunHotel uses a custom on-device Small Language Model (SLM) that runs entirely on the user's device via ONNX Runtime. All voice processing happens locally — no data is sent to the cloud, ensuring both privacy and zero-latency operation even in areas with poor connectivity.
What languages does RunHotel support?
RunHotel supports Hindi, Hinglish (Hindi-English code-switching), and English. The on-device model was trained specifically on hotel management vocabulary in these languages to ensure accurate recognition of room numbers, guest names, and operational commands.
How long did it take to build RunHotel?
The initial MVP was delivered in 8 months with a team of 5 engineers. This included the Flutter mobile app, Next.js dashboard, Node.js backend, and the custom on-device voice model. Ongoing development continues to add new features and expand language support.
