Xetlink Technology Inc has built Xetlink AI, a real-time multilingual communication solution for video conferencing. The technology provides speech-to-text and speech-to-speech translation, meeting transcript analysis, and transcription capabilities, allowing individuals and organizations to conduct meetings with people who speak different languages without the need for a translator. Approximately 55 million meetings occur worldwide every day, and while English serves as a common global language, 75% of the world's population does not speak it.
Approximately 55 million meetings occur worldwide every day, many involving participants in different locations, and while English serves as a common global language, 75% of the world's population does not speak it. Xetlink set out to support meetings in multiple languages in real time, but the product it had was restricted to a single video platform and expensive to run.
Xetlink's solution was restricted to Zoom alone, which capped what it could be sold into. A translation product tied to one video platform reaches only organizations using it, and only for meetings held there. A multinational team, precisely the customer with a language problem, is likely on Teams with one client, Google Meet with another, and Webex internally.
A pivotal aspect was capturing the voice of the speaker in each meeting, which meant a bot that could autonomously join meetings, record conversations, and pass content to a converter. The difficulty is architectural, since conferencing platforms are built so audio stays between participants, which is correct and also exactly the obstacle.
Building Teams, then Google Meet, then Webex as separate projects would have paid for the hard part three times, since joining a meeting, identifying the active speaker, capturing audio, and streaming it out are common to all of them. Adding platforms also multiplies concurrent instances, so coverage and cost had to be solved in one body of work rather than sequentially over three releases.
Automation bots were implemented for Google Meet, Webex, and Microsoft Teams, using Puppeteer to launch browsers and join meetings as participants, streaming audio to a server once the active speaker was identified. Every one of these services runs in a browser, so the browser is the common surface, and one bot that drives it can operate all three.
Puppeteer scripts captured active users, chat text, and the activity of departing participants, transmitting everything over a websocket. Meetings are not only audio, and participants less confident in a second language often type rather than speak. Tracking departures matters too, since a bot unaware a meeting has emptied keeps consuming resources.
Xetlink diversified its service offering across multiple meeting providers, breaking free of its restriction to Zoom alone, and that expansion contributed to an 8 percent rise in overall revenue.
Transitioning from Windows-based automation to a Linux-based solution achieved a substantial 22 percent reduction in operational costs, with Docker-based scaling handling traffic efficiently after.
Premium features such as the ability to simulate bot voices gave Xetlink a notable competitive advantage over industry rivals, further solidifying its position in the multilingual meeting market.
Eternalight built Xetlink's meeting automation on Node.js with PM2 process management, using Puppeteer to drive browsers and join Google Meet, Webex, and Microsoft Teams as a participant. Audio events stream continuously over WebSocket, FFmpeg handles voice simulation for speech-to-speech conversion, and the platform runs on Linux in Docker containers, replacing Windows-based automation.
Brainstorming
Wireframe
Development
Send us a message, and we’ll promptly discuss your project with you.
Once voice was recorded and converted to text, the bot needed to translate it into the user's preferred language. Real time is the hard part of that sentence. A transcript translated afterwards is a useful record, while a translation delivered during the meeting is what lets a conversation happen, and the difference is entirely latency.
Xetlink previously relied on Windows-based automation for meeting participation. This is operational rather than functional, and it binds at scale, since every concurrent meeting needs its own automation instance and the environment cost multiplies with them. A product serving a share of 55 million daily meetings cannot carry that per instance.
A websocket streamed audio events continuously so users at the other end received live translated text in their chosen language. Request and response would mean recording a chunk, sending it, waiting, and receiving a result, with latency accumulating at each step. A persistent connection is what compresses the gap between speaking and reading.
A bot simulating user voices with FFmpeg enabled voice conversion between languages, while migration from Windows to Linux and then Docker made the expansion affordable. Reading a translation keeps a participant a step behind, whereas hearing one lets them stay in the conversation, which is why it became a premium capability.
User Feedback
Launch
Integration
A cross‑platform, open‑source JavaScript runtime environment for building fast, scalable server‑side applications.
Browser automation framework for joining and managing meetings.
Audio processing and voice simulation for multilingual communication.
Real-time streaming of audio, transcripts, and translated content.
Containerized deployment for scalable and efficient operations.
Lightweight server environment reducing infrastructure costs.
Process manager ensuring reliable application uptime and monitoring.
