Find parts by photo or QR code, with similarity-based candidates, a confidence indicator and the option to return no match.
RIZUMU / APPLIED AI
ImageBot
PhotoSimilarityCandidates
ILLUSTRATIVE FLOW · Computer vision for catalogs
The challenge
Identifying a part without knowing its code can mean searching the catalog by hand and relying on the experience of whoever knows each item.
The solution
A visual recognition project that combines catalog images, embeddings and vector search to present candidate parts. History, feedback and abstention are built into the identification flow.
Identification starts with the catalog: each SKU needs images and information that help tell items apart. The project uses DINOv2 to represent visual features and vector search to find candidates by similarity. A photo search can return several options, while a QR code offers another path to identification.
The flow assumes that a single image may not be enough. That is why candidates, confidence and abstention are part of the experience, along with search history and feedback. A commercial deployment should assess the photos, capture conditions and differences between parts in the chosen catalog to decide when to suggest an item and when to ask for review.
What the project brings together
Identification by photo or QR code
Catalog organized by SKU
Candidates, confidence and abstention
Import, history, feedback and API
FROM PROJECT TO OPERATIONS
Use cases
Illustrative scenarios to picture how this work can be applied in a company’s day-to-day operations.
USE CASE 01
A part with an unreadable code
The scenario
A maintenance team needs to look up a part whose code has worn off, but it has photos and a spare-parts catalog.
How to apply it
Compare the image against the selected set and present candidates so the person can check their features before deciding.
What the team gets
A visual search with SKU options, a confidence indicator and a record of the confirmation or the open question.
USE CASE 02
Photo-based support
The scenario
A service center receives images of components that the requester can’t name and needs to organize its internal search.
How to apply it
Make visual search available to support the service agent, linking each search to its history and to feedback on the result shown.
What the team gets
An identification support flow with a searchable catalog and inconclusive cases routed for analysis.
GROUP / Data and catalog
Offers built on ImageBot.
Choose the starting point closest to your challenge. Each offer proposes a scope of work based on the experience of this project.
ImageBot01
Visual catalog by SKU
Who it’s for Companies with parts and images scattered across different sources.
Prepare a subset of the catalog for visual search, defining how items are organized and how photos, codes and the attributes that help verify a match fit together.
Proposed deliverables
Catalog structure with SKUs, categories and images
Import routine for the selected sources
Photo capture guidelines and a reference set for evaluation
Who it’s for Product teams that want to add identification to an existing application.
Design the integration between part identification and a system of your choice, keeping the candidates and the inconclusive-result status in the response.
Proposed deliverables
API contract with input, candidates and response states
Integration with one system flow defined in the scope
Search history and a feedback mechanism for identifications
We can combine offers and adapt the deliverables. Data, integrations, acceptance criteria, timeline and investment are defined in the proposal for your company.