Can we change model partway through?

Yes, and it is common. Engagements often start as a scoped project and move to a dedicated team once the direction is clear. Changing model is a contract amendment, not a restart.

Do you take on work on an existing codebase?

Yes. Most of our work is on systems that already exist and already have users. We start with a short read-through of the code and infrastructure before quoting anything.

What is the smallest engagement you take?

Small enough to be a real test of whether we work well together. Tell us the first slice you would want delivered and we will say whether it is a sensible starting point.

Services

SofinWave works in four registers: team models, engineering services, AI delivery, and the training data underneath it. Which one fits depends on whether you need capacity, specific skills, someone to own an outcome, or the dataset a model will learn from. If you are not sure, describe the problem and we will tell you which we would propose and why.

Team models

These three differ in who owns the outcome. Pick by how much of the delivery responsibility you want to keep.

Offshore development — a complete team that owns delivery of a product or system, with its own process and reporting.
Dedicated team — engineers assigned to you long-term who work inside your process, your board, your standups.
IT staff augmentation — named individuals filling specific gaps in an existing team, managed by you.

Engineering services

These are scoped around a technical outcome rather than a team shape, and are usually delivered by one of the models above.

Custom software development — web, mobile, and backend systems built to your business rules.
System integration — connecting ERP, CRM, payment, and third-party systems into one reliable flow.
DevOps & cloud — CI/CD, infrastructure as code, observability, and cloud cost work.

AI delivery

Three purchases that get conflated because they all involve a model. Each fails in a different way, so each has its own page.

AI implementation — taking a model from a convincing demo to a system with evaluation, cost control, and a rollback path.
LLM integration and RAG — connecting a model to your own data without it inventing the parts it does not know.
AI agents — multi-step workflows that act rather than answer, with the guardrails that makes tolerable.

AI training data

The data an AI system learns from, collected and labelled by the team that then has to train on it. The overview page explains how the four fit together.

Data collection — video, image, audio, document, and scripted scenarios captured to a written specification.
Egocentric and exocentric capture — synchronised first-person and multi-view recording for robotics and embodied AI.
Data annotation — image, video, 3D point cloud, document, and audio labelling.
RLHF and SFT data — demonstrations, preference ranking, evaluation, and red teaming for LLM post-training.

How to choose

The practical test is who will be accountable when a deadline slips. If the answer has to be the vendor, you want offshore development. If your own engineering manager will own it, staff augmentation or a dedicated team is a better fit and usually cheaper.

Frequently asked questions

Can we change model partway through?: Yes, and it is common. Engagements often start as a scoped project and move to a dedicated team once the direction is clear. Changing model is a contract amendment, not a restart.
Do you take on work on an existing codebase?: Yes. Most of our work is on systems that already exist and already have users. We start with a short read-through of the code and infrastructure before quoting anything.
What is the smallest engagement you take?: Small enough to be a real test of whether we work well together. Tell us the first slice you would want delivered and we will say whether it is a sensible starting point.

Tell us what you are building

Send a short description of the system and the constraint you are hitting. We reply within one business day.

Book a consultation