Catalogue Enrichment from Product Images

• Head of E-commerce
• Catalogue Manager
• Marketplace Operations Lead
• D2C Brand Manager

Every new SKU needs a listing on your website, two or three marketplaces and several quick-commerce apps, each with its own template, attribute list and image rules. Catalogue teams type attributes from samples and artwork files, and errors show up as rejected listings or returns.

We build an enrichment pipeline that takes product photos, pack artwork and back-of-pack images and produces a complete, channel-ready listing draft. Vision models identify visual attributes such as colour, pattern, neckline or pack type. OCR and an LLM read pack text for net quantity, MRP, ingredients, veg or non-veg mark, FSSAI licence number, manufacturer details and best-before period. Values are normalised to each channel's allowed list and mapped to its template. Titles and bullets are drafted in your brand voice from facts found on the pack or in your product data only. The catalogue team approves before anything is published.

• Extracts visual attributes such as colour, pattern, sleeve, neckline, material look and pack format
• Reads back-of-pack text for net quantity, MRP, ingredients, allergens, FSSAI licence and manufacturer or importer
• Normalises values to each marketplace's allowed list and fills its flat-file template
• Drafts titles, bullets and descriptions using only facts found on the pack or in product data
• Checks image sets against channel rules for background, count, angles and resolution
• Links every attribute to the image region it came from for quick review

1. 01Collect product assetsPhotos, pack artwork and back-of-pack images arrive with the SKU code from the PIM, a shared drive or the photo studio.
2. 02Extract attributesVision models read visual attributes and OCR with an LLM reads pack text, each value with a confidence score.
3. 03Normalise and mapValues are mapped to each channel's allowed list and template, and missing mandatory fields are listed.
4. 04Draft listing copyTitles and bullets are drafted from verified facts, in the style each channel expects.
5. 05Catalogue reviewThe catalogue executive checks low-confidence values, declarations and copy, then approves the listing.
6. 06Publish and trackApproved listings are exported to each channel, and rejection reasons are fed back into the mapping rules.

1. Asset intake
   • PIM, DAM or shared drive connector
   • SKU and asset grouping
2. Extraction
   • Visual attribute models by category
   • Pack text OCR with LLM field parser
   • Image rule checker
3. Normalisation
   • Channel attribute dictionaries and allowed values
   • Template mapper for marketplace flat files
4. Content and review
   • Copy drafting restricted to verified facts
   • Catalogue review screen with image highlights
5. Publishing
   • Marketplace flat-file and API export
   • Rejection feedback loop

Catalogue executives approve every listing before it is published and confirm any low-confidence attribute. Statutory declarations and any health, ingredient or performance claim are always checked against the pack image by a person.

• PIM or DAM such as Akeneo, Salsify or an in-house system
• Shopify, Magento or your website CMS
• Marketplace seller portals and flat-file templates
• ERP item master
• Shared drives or photo studio storage

• Copy uses only facts present on the pack or in approved product data
• Attribute values restricted to each channel's allowed list
• Every value linked to its source image region
• No listing published without catalogue approval

• Mandatory declarations under the Legal Metrology (Packaged Commodities) Rules 2011 are extracted and checked, and the catalogue team confirms them before publishing.
• Food listings carry FSSAI labelling details, such as the licence number and veg or non-veg mark, exactly as printed on the pack.

• Retail, D2C & Quick Commerce

  Shelf Availability and Planogram Compliance

  Turn merchandiser and store shelf photos into SKU-level facings, out-of-stock gaps, price-label checks and planogram scores, with doubtful detections sent to a reviewer.

• Retail, D2C & Quick Commerce

  Review and UGC Insights for Product Teams

  Classify reviews, Q&A, social comments and return reasons into product issues by SKU and batch, including Hinglish text, and give product teams a weekly digest with source quotes.

• Retail, D2C & Quick Commerce

  Guardrail Testing for AI Shopping Assistants

  Adversarial testing of AI shopping assistants for unauthorised discounts, wrong prices, order-data leakage and unsafe product advice, with validated findings, fixes and a retest.

• Automotive & Mobility

  Spare Parts Catalogue Matching from Photos

  Identify spare parts from customer or mechanic photos by reading part numbers, matching catalogue images and filtering by vehicle fitment, then confirm stock and price before quoting.