We build and maintain software and the machines it runs on.
Contrivance Inc. builds and runs multimodal AI systems: the models, the software around them, and the servers they sit on. We do it for our own products first, which is the reason we can do it for yours.
What we do
Multimodal AI systems
Software using artificial intelligence to process, analyse and generate video, image, audio and text. Transcription, visual understanding, retrieval, titling, summarisation, recommendation, built to answer from your material rather than from a general index.
Software
Native iOS and Android applications, the backends that serve them, and the APIs in between. Real-time features included from the start: messaging, calling, live data.
Inference infrastructure
The machines as well as the software: GPU servers and workstations specified, assembled and configured for the workload; model serving, capacity planning and the operational work of keeping them up; and software optimised for the hardware it runs on. Self-hosted by default, so cost and latency stay under your control instead of a vendor's.
Research & consultation
Research and development, and consultation related to it, in artificial intelligence, machine learning and multimodal neural network systems. Model selection, evaluation harnesses, and honest numbers on what a system will cost to run before you commit to building it.
What we've built
Everything above we built for our own product first. That is the portfolio: a system in daily use, not a deck.
- ProductCohlta, a video platform for cities, native on iOS and Android, with a location-aware feed, a live event map, and an AI assistant that answers from the footage. cohlta.app
- AIA self-hosted multimodal stack: speech-to-text, visual analysis, embeddings and retrieval, title and summary generation. Running on GPUs we own, with no third-party AI API in the path.
- Real-timeDirect messaging and voice calling, end-to-end encrypted, over WebRTC with native call integration on both platforms.
- GeospatialA live event map assembled from crawled and geocoded sources, with clustering, filtering and area-scoped question answering.
- InfraThe inference servers underneath all of it: model serving, queueing, capacity measurement and the cost model that decides what runs where.
Where to reach us
Tell us what you're trying to build and what it has to run on. If it's a fit we'll say so, and if it isn't we'll say that too.
[email protected]