sistema labs

a research corporation

Intro

We are a research corporation that seeks to develop novel applications of the transformer architecture to problems outside NLP.

The decoder-only transformer variant has proven very effective in sequence modelling for NLP applications, giving us LLMs such as the GPT family and Claude. However, there are very many problems that can be addressed with sequence prediction or autoregression models, in fields as varied as robotics and logistics, economics and finance, the biological and earth sciences, and various fields in computer science itself.

The cost of evaluating this approach can be quite high in terms of researcher time, and depending on the size of the model and the data, also in terms of compute. We developed the sequifier package to push down the cost of this initial evaluation, and enable the discovery of those applications that are amenable to modelling with transformers. Crucially, it allows for multi-input multi-output sequence prediction and autoregression, thereby covering a much larger space of possible applications.

Reach out if...

You should definitely reach out to us if:

You have some modelling problem that can be framed as a
- sequence classification
- sequence regression
- sequence autoregression on labels or real values
You have data for this problem that you would be able to share
You want to cooperate on exploring this modelling problem with a transformer model

...or alternatively, if you just want to discuss any aspect of transformer applications to non-NLP problems :)

Where to go from here

github: explore the sequifier package that enables the quick configuration, training and inference of many-to-many transformer models
blog: read about specific case studies & results
learnings: summaries of our learnings from applying transformer models in various domains

Impressum

Working on it... in the meantime, this is my personal website.

Last updated: