About us
Research · Engineering · Technology
ORCNEIT is a Russian research and engineering company. We work on our own technologies in artificial intelligence, neural engineering and software.
Developing a model includes everything needed to train it, test it and prepare it for use.
We consider the model, data and software tools together: a change to one part affects the others.
Our main project is ORCNEITGPT. Alongside the model, we are building tools for training, preparing data and evaluating results.
These tasks depend on one another: a research idea needs to be tested, testing requires data and software, and training needs computing resources.
01
Our work ranges from studying language models to building the software and computing systems they run on.
These areas come together in the work on a single model.
These are connected areas of work, not a list of separate products. For example, data preparation is needed both to train the model and to test its responses.
Alongside the model architecture, we need data, computing resources and training software. We also need tools to monitor training, test the model and prepare it for use.
02
An idea becomes useful for development when it can be tested.
In an experiment, we need to know which question we are asking, what we are measuring and which conclusions the results support. This helps distinguish a tested hypothesis from an assumption.
The ORCNEIT working cycle
Idea
Research
Experiment
Measurement
Engineering implementation
Next experiment
Results should be documented so that we can revisit them and repeat the test. This applies to changes in both the model and the software tools.
A successful experiment can inform the next change. If a result does not support the idea, that also needs to inform further work.
Key area
The language model that forms the starting point for ORCNEIT’s future tools.
The first priority is training the model and evaluating its language capabilities.
In parallel, we are developing the software and computing infrastructure needed for this work.
We then want to build an assistant around it for work with text, code, documents and external services.
The ORCNEITGPT page describes these plans and their limitations in detail. They are not a list of features already released.
03
We develop the research and software foundations of our projects within ORCNEIT.
This work covers the model, training and data processing tools, storage systems and monitoring tools.
For us, independence means being able to understand key components, change them and continue their development.
We do not use third-party ready-made solutions or another company's technology stack as the foundation of our research and development.
Model training and experiments require computing resources, data storage and software to work with them.
We are developing our own infrastructure for these tasks: data preparation, training and software testing.
It should allow us to revisit earlier tests and use their results in subsequent experiments.
Public descriptions exclude details that could compromise infrastructure security, such as protected system configurations, access credentials and other internal technical information.
04
What guides our development work and evaluation of results.
Develop key technologies within the company and retain the ability to adapt them to future projects.
Assess changes using data and test results. A good impression from a single model response is not enough.
Use earlier experimental results to guide the next steps, taking the limitations we find into account.
Record the conditions and results of an experiment so that the test can be repeated.
Understand how the system’s key components work and how changes affect its behaviour.
Distinguish information useful to a public account from internal details related to security and intellectual property.
05
ORCNEITGPT is our main project, but the company’s goals extend beyond it. We want to use the knowledge, software and infrastructure from this work in future research.
That requires tools we can refine and apply to new tasks, as well as results that we can understand and build on.
Our areas of work may expand. We want to retain the ability to develop and test key technologies independently.
Developing technologies through our own knowledge and tools.
Work on the first model should yield more than the model itself: research experience, software and ways to evaluate results.
Developing these further is part of ORCNEIT’s long-term work.
06
ООО «ОИЦ НИНТ» / ORCNEIT LLC
A Russian research and engineering company.