How We Work

Every IntelliSensei engagement follows the same four stages. The stages are fixed; the duration and commercial shape of each flex to fit the project. This page describes what to expect so there are no surprises after the first call.

1. Discovery

Typically one to two weeks. We start by understanding the problem you are trying to solve, the data you have, the constraints you operate under (latency, cost, compliance, hardware) and how success will be measured. For existing systems this includes a technical review of your code, training pipeline and deployment. The output is a written assessment with a recommended approach, an honest view of risk, and a scoped proposal for the next stage. If the honest answer is that you do not need a PyTorch consultancy, we say so here.

2. Pilot

Typically two to six weeks. Before committing to a full build we prove the approach on a bounded slice: a fine-tuned model on a sample of your data, a retrieval pipeline over one document set, a torch.compile benchmark on your production model, an ExecuTorch export running on one target device. Pilots end with measured results against the success criteria agreed in discovery, and a go/no-go recommendation. A pilot that fails cheaply is a good outcome.

3. Build

Typically one to four months, depending on scope. The production build: data pipelines, training or fine-tuning runs, evaluation harnesses, serving infrastructure, integration into your systems, and documentation. We work in your repositories, on your cloud accounts, in short iterations with a demo at the end of each. Everything we build is yours, with no proprietary runtime or lock-in.

4. Support

Ongoing, sized to your needs. Models drift, dependencies move, and PyTorch releases keep coming. After launch we offer monitoring and maintenance retainers, scheduled upgrade work, and on-call support for production incidents. See our support and maintenance service for the options.

Engagement shapes

Project-based. Discovery, pilot and build are scoped and priced as milestones with defined deliverables. This is the right shape when the outcome is clear and you want a fixed commitment.

Time and materials. When requirements are genuinely still moving, we work at a weekly rate against a prioritized backlog you control, with written status every week.

Staff augmentation. A senior PyTorch engineer joins your team for a defined period, working in your process and tools. This is the right shape when you have the direction and need the hands. All our engineers are US-based and available for remote or onsite work.

Training and advisory. Short, focused engagements: a personalized PyTorch training course for your team, an architecture review, or a strategy engagement ahead of a build decision.

What we expect from you

A technical point of contact who can make decisions, access to the data and systems in scope from day one, and candour about constraints. Engagements go fastest when we can talk directly to the people who will own the system afterwards.

What you can expect from us

Senior engineers only, written deliverables at every stage, measured results rather than assertions, and a straight answer when something will not work.

Ready to talk about your project? Contact us to set up a discovery call.