Frequently Asked Questions

Answers to the questions we hear most often on first calls. If yours is not here, contact us and we will answer it directly.

Do you work with LLMs and generative AI?

Yes. A large share of our current work is LLM engineering in PyTorch: fine-tuning open-weight models on client data, optimizing inference with vLLM and quantization, and building retrieval-augmented generation systems. We also still do a great deal of classical deep learning (computer vision, forecasting, recommenders), which is often the better tool.

Do you only work in PyTorch?

PyTorch is our specialty and the reason clients come to us. We regularly work with the ecosystem built on it (Hugging Face, vLLM, ExecuTorch, Lightning, Triton) and we modernize legacy TensorFlow and JAX codebases onto PyTorch 2.x. We do not take on projects whose core is a different framework.

Are your engineers in the United States?

Yes. All IntelliSensei consultants are located in the US and work in US time zones. Remote is our default; onsite work is available.

What is the minimum engagement?

Our smallest engagement is a discovery or advisory block of roughly one to two weeks. Most projects are larger, but we would rather start small and prove value than oversell scope. See how we work for the stages.

Can you rescue a stalled or failing project?

Yes, and it is one of the things we do most often. Typical rescues: a model that performs in notebooks but not in production, a training pipeline nobody can reproduce, a deployment on unmaintained software such as TorchServe, or a PyTorch 1.x codebase that cannot be upgraded. We start with a short technical review and a written plan, then fix it.

Do you provide staff augmentation?

Yes. A senior PyTorch engineer can join your team for a defined period under your process and tooling. Contract-to-hire is available for long-term roles.

Who owns the code and the models?

You do. Everything we produce during an engagement, including code, trained weights, adapters, datasets and documentation, is delivered into your repositories and accounts. We do not use proprietary runtimes.

How do you handle our data?

Work happens in your cloud accounts or on your hardware wherever possible. When data must be handled outside your environment we agree the controls in writing beforehand. Fine-tuning open-weight models on your own infrastructure is, in practice, the most private option available for LLM work.

Do you offer training for our team?

Yes. Personalized PyTorch training is delivered as hands-on workshops tailored to your codebase and goals, from PyTorch fundamentals to distributed training and LLM fine-tuning.

How do we get started?

Contact us with a short description of your project. We will reply within one business day to set up a discovery call.