Search
Catalogue retrieval
Hybrid search, facets, a merchandising pin, evals on a labelled set.
Retail & E-commerce AI
Search, recommendations, content, and service: retrieval and agents against catalogue reality, margin, and a merchandising veto.
Created by Baljeet Dogra
Course Objectives
- Build retrieval over a messy catalogue, not a clean demo JSON.
- Respect merchandising rules as code, not as a prompt hope.
- Evaluate with ranking and business metrics, not only BLEU-shaped vanity.
- Know when a recommendation is a ranking problem, not an agent.
What you will produce
Rule
Veto path
A generated description or rec that cannot violate a rule you encoded.
4 weeks curriculum
Expand a part for the syllabus. Content stays searchable when closed.
01 Catalogue truth Week 1
SKUs, variants, stale prices, attributes nobody filled in.
- Catalogue
- Attributes
- Staleness
- Identity
02 Find and rank Week 2
Search, recs, hybrid, business features in the ranker.
- Search
- Recs
- Hybrid
- Features
03 Generate under rules Week 3
Copy, attributes, service answers. The veto is a test.
- Content
- Service
- Rules
- Evals
04 Online sense Week 4
A/B caution for recs, capstone with a merchandising veto.
- Online eval
- Margin
- Veto
- Capstone
Who this is for
Engineers in retail and marketplaces
You have a catalogue. A generic RAG tutorial will not survive it.
Ecommerce tech leads
Merchandising will not accept “the model said so”.
Prerequisites
- Python and APIs
- You have seen a product catalogue or storefront
Not a fit if
- Generic ChatGPT-for-marketing
- People with no interest in ranking or rules
Related: Vector Database Deep Dive · Hands-on LangChain
Questions
Shopify or SFCC?
Catalogue-shaped, platform-agnostic. Salesforce Commerce appears as context, not as a lock-in.
Will we train a recsys from scratch?
Enough ranking to be dangerous. Deep recsys is a follow-on, not this month.
Ready to start?
Four weeks to catalogue search with a merchandising veto. Create an account to enrol.
Enrol now