MeetSolanyum
A low-latency, AI-powered video conferencing platform.
<150ms
Avg Latency
50+
Concurrent Participants
98%
Transcription Accuracy
E2E
Encryption
Project Description
MeetSolanyum is a real-time communication platform, engineered end to end with meticulous attention to detail, that blends a low-latency video conferencing experience with AI-driven audio and video processing.
Media infrastructure
- To avoid the exponential bandwidth cost that a traditional P2P mesh architecture incurs as participant count grows, the system is built on an SFU (Selective Forwarding Unit) media architecture: each participant sends only a single stream, which the server intelligently forwards to the relevant peers, letting dozens of people join the same call without any degradation in performance.
- A comprehensive NAT traversal layer, backed by STUN and TURN servers, guarantees that connections are established reliably no matter how restrictive a participant's network configuration is.
- Adaptive bitrate mechanisms and simulcast support ensure that even users on weak internet connections stay in the meeting seamlessly, without dropping out.
Accessibility and privacy
- The platform's live speech recognition engine transcribes conversations in real time with high accuracy and generates instant captions, pushing accessibility to its limit — these captions can also be automatically turned into shareable, searchable written transcripts once the call has ended.
- End-to-end encryption (E2EE) guarantees that the content of every call stays strictly between its participants, with no third party — including the server operator — ever able to access the media stream, making the platform safe to use in scenarios that demand full privacy, such as healthcare, legal, and enterprise advisory consultations.
Scalability
- A region-based media server clustering architecture automatically routes every user to the geographically nearest, least-loaded server.
- This minimizes end-to-end latency while allowing the system to scale horizontally without friction as traffic grows.
The result is a smooth, dependable communication experience where the technology fades into the background and users can focus purely on the conversation.
Technologies Used
Key Features
- 1 WebRTC-based SFU (Selective Forwarding Unit) media streaming delivers low latency and high stability even in multi-participant calls; each participant sends only a single stream, which the server intelligently forwards to relevant peers, keeping bandwidth usage from scaling exponentially with participant count.
- 2 Real-time speech recognition and live captioning make calls more accessible for hearing-impaired participants and teams speaking different languages, while also enabling shareable, searchable written records to be generated after the call.
- 3 End-to-end encrypted data transmission (E2EE) keeps call content strictly between participants; an additional key exchange layer means even the server side cannot access the media stream's content, making it suitable for scenarios requiring full privacy such as healthcare or legal consultations.
- 4 Adaptive bitrate and low-bandwidth optimization keep even users with weak connections seamlessly in the call; WebRTC's built-in bandwidth estimation mechanisms automatically adjust video quality based on packet loss and latency variation, aiming to keep the call from ever dropping.
The client experienced serious lag and freezing in calls with more than three participants because the system relied on a pure P2P mesh architecture, where each participant sent data separately to everyone else — exponentially consuming bandwidth and dropping low-bandwidth users from calls.
We migrated to an SFU (Selective Forwarding Unit) architecture where each participant sends a single stream that the server forwards to relevant peers. Adaptive bitrate and simulcast support keep even low-bandwidth users seamlessly connected.
Impact & Results
Showcase & Interface
