About — Smart Waves LLC

What you are
actually hiring.

Smart Waves is a computer vision and data consultancy in Dallas, Texas. The people who write the models are the people you talk to. There is no layer in between.

MOVEMENT OBJECTS DEPTH FACE WAVEFORM DOCUMENTS FORECAST ROUTING Eight of the things these systems produce

01Practice

Twenty-five years, most of it close to the metal.

Smart Waves is built on engineering that started well before machine learning had a marketing department. The last stretch of it has been computer vision: systems that watch real rooms, with real light and real people moving through them, on hardware that has to hold its frame rate all day without anyone watching.

That is the part nobody demos. A model that works on a clean test set and a system that survives a Monday morning are different achievements, and the second one is where the money and the difficulty both live.

The practice is deliberately built as a team — engineering, data science, and project management — because a specialist shop that depends on one person is a risk the client ends up carrying.

02Difference

We design the networks. Not just the training scripts.

Most firms offering computer vision are fine-tuning something they downloaded. That is legitimate work and often the right answer — it is simply not the same skill, and the difference only shows up when something breaks.

Smart Waves builds the architectures. Dataset collection and labelling, training, evaluation, and export to run in real time on the device it will actually live on.

The reason it matters is failure. When an off-the-shelf model falls over on your data, a team that only fine-tunes has run out of moves. Knowing why it failed, and being able to change the thing that failed, is what you are paying for.

Pose detection — custom keypoints Colour classification Face recognition Object detection — trained, not borrowed And others

03Method

Small engagements. Honest numbers.

01You talk to the engineer
The engineers who scope the work are the engineers who build it. Project management keeps the work on track; it does not sit between you and the people doing it.
02Evidence before budget
Candidate approaches get tested on your data, not on a benchmark that flatters everyone equally, before anything is committed.
03Failure cases first
You will hear where a model breaks before you hear what it scores. A number without its failure modes is not information.
04It has to run in production
A notebook that works is not a system that works. If it cannot hold up on your hardware, on your worst day, it is not finished.

04Fit

Worth a conversation. And when it is not.

Call us when

  • The footage already exists and nobody has time to watch it.
  • A downloaded model got you most of the way and then stopped.
  • It has to run on the device, in the room, without a connection.
  • The documents are in two scripts, or half of them are handwritten.
  • The data is spread across systems that disagree with each other.
  • You want the pipeline, the model, and the reporting on top of it from one team.
  • Someone has told you it cannot be done, and you are not convinced.

We push back when

  • Nobody can yet say what success would look like. We would rather define it with you than quote against a guess.
  • The data does not exist yet and there is no plan to collect it.
  • The decision has already been made and what is wanted is agreement.
  • The deadline is next week. Testing on your data takes longer than that.
  • The budget assumes the first model will be the last one. It never is.

The second list is not a refusal. Every item on it is fixable, usually inside one conversation, and all of them are cheaper to fix now than four months in.

Bring us the hard half of the problem.

A feed nobody has time to watch, a pile of documents nobody has time to read, or a warehouse nobody trusts.

Start a conversation →