Self-Ready overview

Case study · Product Strategy

Self-Ready

Turning product imagery into device-ready recognition data.

Role
Lead Product Designer
Contribution
End-to-end product design
Focus
UX strategy, information architecture, user journeys, interaction design, UI design, prototyping and developer collaboration
Platform
Retail image intelligence and product readiness
The Self-Ready Workbench: ingestion, processing, review, product data and device readiness in one place.

How might we make the process of preparing and validating product imagery clear enough for operational teams to manage, while maintaining the quality of data used by the device model?

03 The design objective

Create a connected workflow from image intake to device-ready product data.

04 Design approach

My approach

  1. Two distinct entry points: Manual Bulk Ingestion and Scheduled Image Capture.
    Ingestion History: submitted batches with successful and failed image counts.
  2. Capture Sessions: processing states, image counts, identified products, facings, AI quality, labels read and barcodes.
  3. Image Review: the product crop, its shelf label, AI confidence and reason, and a direct Confirm or Reject.
  4. Blocklist: exceptions with Needs Review, Confirmed and Rejected states, linked to the original capture session.
  5. Product Database: confirmed product information and its imagery.
    Product details: Active, Seasonal and Inactive imagery.
  6. On-device Index: the product data prepared for device use.
Self-Ready end-to-end user flow: 1. Bring in imagery — Open Self-Ready Workbench, Choose manual bulk ingestion or scheduled capture, Submit image batch / capture request, Track batch in Ingestion History; 2. Process & review — Open Capture Session, Review processing and image results, Inspect product crop and shelf-label match, Confirm or reject AI association; 3. Improve catalogue — Resolve unmatched products / barcodes, Send rejected cases to Blocklist when appropriate, Review Blocklist items and resolve, Maintain confirmed Product Database records; 4. Prepare device model — Manage active product images, Maintain seasonal and inactive image states, Refresh the On-device Index, Use prepared product data to train / update device model
Fig. 1 Conceptual journey from documented Self-Ready UX; exact model-training mechanics and index refresh triggers require engineering confirmation.
Self-Ready product architecture: Input layer — Manual Bulk Ingestion, Scheduled Image Capture, Ingestion History, Capture Sessions; Intelligence & decision — Image processing / detections, Product and shelf-label association, FDE Image Review, Blocklist review and resolution; Product data layer — Confirmed Product Database, Active image set, Seasonal image set, Inactive / historical images; Device readiness — On-device Index, Eligible active product imagery, Device model training / update, Device-ready recognition data
Fig. 2 Conceptual product architecture. It represents UX and information flow, not unverified model-training pipelines or backend services.

07 Decisions

Key design decisions

  1. 01

    Separate ingestion methods

    Keep manual bulk ingestion and scheduled image capture as distinct workflows.

  2. 02

    Decision-focused image review

    Make product-label validation the primary task instead of exposing technical pipeline details.

  3. 03

    Preserve decision history

    Keep rejection, blocklist and review states distinct so users can understand what happened to an item.

  4. 04

    Manage image lifecycle explicitly

    Separate active, seasonal and inactive imagery rather than treating all product images equally.

  5. 05

    Connect product data to device readiness

    Make the relationship between confirmed product imagery, the On-device Index and device model preparation understandable.

Intro film. Hello World. I’m Nishanth, Lead Product Designer. I make things that help others do their thing. Even when it’s complex, I make it clear. +15% Best engagement lift, 5+ Years designing products, 04 Design systems built or shaped. I design it in Figma, then ship it in code.