DYNAMO (Dynamo Tech Solutions) has built its own online meeting system and put it into daily use across the company. By adopting LiveKit for real-time media and Cloudflare Zero Trust for access control, we avoided hand-building the two most expensive parts of such a system — video delivery and authentication. Recording, transcription and minutes are then automated with WhisperX and a locally hosted AI model, so that meeting audio never has to be handed to an external transcription service.

Why we built it

Mature online meeting services already exist, so building our own needed a reason. There were two. First, we wanted meeting recordings and transcripts to stay under our own control. Second, we believe the most reliable way to sharpen the AI development and operations expertise we offer clients is to run an AI-powered product ourselves, every day, in real work.

At the same time, an internal tool cannot absorb months of development. Our design principle was therefore simple: do not build what you do not have to build. We combined proven managed services with open source, and spent our engineering effort only on the parts that genuinely had to be ours.

How it works

1. Domain-restricted Google sign-in — authentication we did not write

Cloudflare Zero Trust Access sits in front of the application, requiring a Google sign-in and admitting only users from approved domains. The authenticated identity is passed through to the application as a request header, which means no login screen, no session management and no password storage had to be implemented in the application itself.

Authentication is both costly to build and maintain, and unforgiving: a small gap in the implementation becomes a security incident. Delegating it to a proven managed service shortened the build and reduced the attack surface at the same time. Administrative functions are protected by a second layer on top of Access, restricted to an explicit allowlist of addresses.

2. Real-time audio and video powered by LiveKit

Connecting participants across locations in real time is demanding work if WebRTC is assembled from scratch. We chose to self-host LiveKit, an open-source platform, and implemented the core of the product — audio and video delivery, screen sharing and participant management — in a fraction of the time.

Features such as the pre-join lobby (camera preview and microphone/device checks before entering), focused display for shared screens, and the participant list have been added and refined in the order they were actually needed, while the team used the system for real meetings.

3. Recording, transcription and minutes automated with local AI

Meeting recordings are written straight to object storage (Cloudflare R2). When the meeting ends, post-processing starts automatically: the audio is transcribed with WhisperX, and a locally running LLM produces the minutes — key points, decisions and action items.

The essential point is that transcription and summarisation never leave our own environment for an external AI API. Confidential discussions stay in-house, while the clerical work of writing minutes is taken care of automatically.

Technology stack

What this means for our clients

We built this system for ourselves, but the design decisions behind it — delegating access control to a managed identity platform, using open-source infrastructure to compress the cost of core features, and processing confidential data with local AI so it never leaves the environment — apply directly to the systems we build and operate for our clients.

We welcome enquiries on cloud infrastructure optimisation, security-first architecture, and automating business processes with AI. Please get in touch.