Image intake and rotation
Accept photographs, identify orientation and preserve the untouched original alongside proposed processing results.
Card Scanner API
Exploring how a stack of card photographs can become organized, reviewable records while preserving image quality and acknowledging uncertain identification.
In development · No public production API or working public demo claimedCard Scanner APIIn-development computer vision and identification workflow · No public production release
FUNCTIONAL SCOPE
What the product is designed to do.
Multiple-card detection
Identify individual card boundaries even when several cards share a photo or appear at different angles.
Front/back image pairing
Associate matching images so card number and set information from the back can support recognition.
Structured identification
Propose player or character, year, product, set, number and parallel with confidence checks.
Review-before-save
Present cropped images and editable metadata, including an explicit unknown state when evidence is insufficient.
ENGINEERING APPROACH
The workflow behind the interface.
01
Detect
Find cards and correct orientation before extracting individual image areas.
02
Pair
Link front and back photos where there is enough supporting evidence.
03
Identify
Extract and compare metadata using suitable catalogs or allowed reference providers.
04
Review
Let a person inspect, edit or reject the proposed result before it is saved.
REAL LESSONS
What the hard cases taught me.
A perfect-looking crop can be wrong
Card edges, corners and rotated photographs matter to collectors. The original image must remain accessible so a proposed crop can be corrected.
Repeated guesses are not independent results
A batch processor should track each detected card separately and reject reused AI output that cannot be matched to the image.
Pricing claims need matching sales
A listing price is not a verified sold price, and an approximate card identity is not an exact match. The workflow should preserve that distinction.
Current status: The scanner remains in development. There is no claim of an available public API, guaranteed recognition accuracy, verified market-pricing coverage or a published developer SDK.
Other project files
CASE STUDYRubyCollect
A real multi-seller trading-card marketplace where listing, collecting, paying, and fulfilling an order have to work together—not as disconnected website features.
Read case study →CASE STUDYDrawMarkAn owner-controlled renovation record designed to make estimates, revisions, decisions and evidence easier to understand and trace.
Read case study →