Prototype
Omni
Ceiling microphones, local AI and a shared team memory
Could an AI share the context of a whole team? Omni was our experiment in bringing that idea into the innovation workspace at Hypersolid / Triple: following discussions, remembering earlier ideas and contributing to the conversation.
Microphones in the ceiling
We fitted the innovation room with ceiling microphones so Omni could follow conversations across the team. A local computer processed the audio, and a shared screen gave its responses a place in the workspace. The idea was to give Omni the context of projects and team meetings as they happened.
From conversation to context
Speech was transcribed in real time and stored in a local conversation database. Language models running on office hardware could then work with that growing record of the team's discussions.
Live transcription
Room audio becomes a running text record of the conversation.
Local memory
Transcripts collect in a database on the office network.
Local language models
Open-source models use that context to contribute ideas and suggestions.
The original setup used DeepSeek R1 on a network of PCs with NVIDIA 4090 graphics cards. We kept trying new open models, with each change affecting how Omni responded and what it could contribute.
Bring earlier ideas back
Omni could point out a concept the team had explored before or suggest an approach we could reuse. Its shared display surfaced these connections during the conversation, giving the group a way to revisit earlier work.
