[ What isTERNARY? ]
[ TERNARY is an Estonian AI and deep technology company. We develop and deploy AI models, speech technology, local AI infrastructure and the software around them. We start from the task and help choose the model, the infrastructure and the technical solution that suit it. ]
[ MODELS ]
Choosing and adapting existing models, and where needed training smaller specialised models from scratch. Fine-tuning, quantisation and optimisation according to the task, the data and the hardware.
[ MACHINES ]
Local AI workstations and in-house servers, sized for the model in use, the load and the way the data has to be handled.
[ SOFTWARE ]
Web, mobile and desktop applications, APIs and integrations that connect an AI model, or other automation, to an existing workflow.
[ LabTERNARY ]
[ TERNARY Lab is our development and testing environment, where we build and develop AI models, speech technology, and the tools and methods needed to use them. ]
[ HÄÄL ]
HÄÄL is a working Estonian text-to-speech (TTS) solution built around a small, resource-efficient model. Development is focused on the balance between speech quality, stability, voice adaptation and low computational cost. It can be used on a server, and locally where the hardware allows.
[ KUULA ]
KUULA is the follow-on project to HÄÄL and an Estonian speech-to-text (STT) solution. Its task is to turn speech into text and to form the input layer of a system where speech recognition, a language model and speech synthesis work together. We are building a connecting layer between them, so that the chain KUULA → LLM → HÄÄL can later be used in different applications.
[ MATRIX ]
Architecture and runtime for compact ternary AI models. The first complete model, MATRIX EST, focuses on the Estonian language, problem solving and local use.
[ Services ]
[ We start from the task and the need, not from a technology chosen in advance. We help establish the right model, the right compute and the right software solution, and assess how they fit together with the existing system. ]
[ Model development ]
We assess whether the task suits an existing model, an adapted one, or training a smaller specialised model from scratch. We work on training, post-training, quantisation, optimisation and deployment.
[ Data and corpora ]
Building, cleaning, annotating and quality-checking training and evaluation datasets and corpora. Where needed, we prepare the data for training or evaluating a particular model.
[ Local AI systems ]
AI workstations and in-house servers for running and training models locally. We choose the hardware according to the model, the load and the GPU and memory required.
[ AI software ]
Web, mobile and desktop applications, APIs and integrations that connect AI models to existing systems and workflows.
[ Local when possible.Hybrid when useful.Cloud when justified. ]
[ Local ]
Suits solutions where the data has to stay under the company's control, where independence from an external service matters, or where a steady load justifies hardware of your own.
[ Hybrid ]
Part of the system runs on the company's own infrastructure and part in an external service. The cloud can be used for larger models or additional compute, while local control is kept where it is needed.
[ Cloud ]
Suits cases where the load varies, where building your own infrastructure is not justified, or where the compute and models needed are more sensibly used as a service.