MuskarTech AI Lab
Teaching software to see what you mean.
The AI Lab is the research side of MuskarTech: applied machine learning, studied carefully and shipped only when it holds up. Our current question: can software find the exact product in an ordinary photo?
Current project
Visual product search
You see a chair in a café, a lamp in a friend’s flat or a spare part on a shelf. You do not know its name, but you have a photo. Visual product search starts from that photo and returns the product, or the closest matches in a catalogue.
It is designed for the photos people actually take: odd angles, busy backgrounds, imperfect light. The model is in closed beta.
Where it can help
Search by photo
Let shoppers search a catalogue with a picture instead of guessing keywords.
Catalogue matching
Match product photos from suppliers or marketplaces to the items you already list.
Inventory lookup
Identify an item in a store or warehouse from a quick phone photo.
How it works
Three steps from photo to product
- 01
Detect
Find the product in the frame and separate it from everything around it.
- 02
Describe
Turn what the model sees into a compact visual representation of the product.
- 03
Match
Search the catalogue for the nearest items and rank them by similarity.
Focus areas
What the lab works on
- 01
Computer vision
Detection and recognition on real-world photos, where conditions are never ideal.
- 02
Retrieval and embeddings
Representations that make “find me something like this” fast and accurate across large catalogues.
- 03
Efficient inference
Making models fast and affordable enough to run in production, not only in a demo.
- 04
Applied language models
Using LLMs where they remove real work in our products and in client projects.
How we work
Useful over impressive
A model earns its place by solving a real task, not by a benchmark.
Measured before shipped
Every model gets an evaluation set before it reaches users.
Show the reasoning
We return matches with their scores, so people can judge the result themselves.
Join the visual search beta
Have a catalogue, a marketplace or an inventory problem that starts with a photo? Tell us about it.