E-commerce
Product content generation across a catalogue nobody could keep current
Placeholder case study. Draft descriptions, attributes and translations generated from supplier data, with merchandiser approval before anything reaches the storefront.
- TypeScript
- Next.js
- Node.js
- PostgreSQL
- pgvector
- Algolia
- AWS
Sample case study — illustrative, not a delivered client engagement
Harbor Retail Group is an invented company used to demonstrate our case-study format and our approach to this kind of work. The technical detail reflects how we would genuinely deliver it; the client name is fictional and the figures under Results are design targets rather than measured outcomes.
Client
Harbor Retail Group
Illustrative multi-brand retailer — PLACEHOLDER CONTENT
- Industry
- E-commerce
- Services
- AI & AutomationWeb & Mobile
- Year
- 2026
The business challenge
PLACEHOLDER — replace with the real engagement narrative. The illustrative situation: tens of thousands of products arrived from suppliers with inconsistent, often unusable copy. Merchandisers rewrote what they could and the rest went live thin, which hurt both search visibility and conversion. Adding languages multiplied the problem.
Objectives
- PLACEHOLDER — Give every product usable, consistent copy
- PLACEHOLDER — Keep a merchandiser approval step before publication
- PLACEHOLDER — Support additional languages without multiplying the workload
- PLACEHOLDER — Improve on-site search through better structured attributes
Discovery and strategy
What we did before writing code
- PLACEHOLDER — Audited catalogue completeness by category and supplier
- PLACEHOLDER — Measured conversion difference between rich and thin listings
- PLACEHOLDER — Interviewed merchandisers about where their time actually goes
- PLACEHOLDER — Agreed nothing publishes without a human approving it
Solution
What we built
PLACEHOLDER — Attribute extraction
Replace with real detail. Structured attributes pulled from supplier feeds and specification documents, normalised to the internal taxonomy.
PLACEHOLDER — Draft copy generation
Replace with real detail. Descriptions generated from verified attributes only, so the copy cannot invent product claims.
PLACEHOLDER — Merchandiser review queue
Replace with real detail. Bulk review with inline editing, prioritised by catalogue impact.
PLACEHOLDER — Translation workflow
Replace with real detail. Additional locales generated from the approved source copy and reviewed per market.
PLACEHOLDER — Search improvement
Replace with real detail. Better attributes feed faceting and semantic search, improving discovery on long-tail products.
Architecture
How it fits together
PLACEHOLDER — Catalogue enrichment pipeline
Replace with the real architecture. The illustrative shape: generation grounded strictly in verified supplier attributes, with approval as a required gate before publication.
- Supplier feedsRaw product data and documents
- ExtractionAttributes normalised to the taxonomy
- GenerationCopy drafted from verified attributes only
- Review queueMerchandiser approval, required
- Product catalogueSystem of record
- Search indexFaceting and semantic discovery
Design approach
Decisions in the interface
- PLACEHOLDER — Built review for volume: keyboard-driven, minimal chrome
- PLACEHOLDER — Showed the source attribute behind every generated claim
- PLACEHOLDER — Sorted the queue by revenue impact rather than arrival order
Development process
How the work ran
- PLACEHOLDER — Evaluation set of merchandiser-written copy used as the quality bar
- PLACEHOLDER — Generation constrained to verified attributes to prevent invented claims
- PLACEHOLDER — Rolled out one category at a time, measuring conversion at each step
Technology stack
What it runs on
- TypeScript
- Next.js
- Node.js
- PostgreSQL
- pgvector
- Algolia
- AWS
Challenges resolved
What went wrong, and what we did
Every project has these. A case study that omits them is a brochure.
PLACEHOLDER — Generated copy invented specifications
Replace with real detail. Restricting generation to verified attributes, and validating output against them, removed the class of error entirely.
PLACEHOLDER — Review became the new bottleneck
Replace with real detail. Bulk approval for high-confidence categories kept individual review for the cases that genuinely needed it.
Results
Objectives and design targets
Because this is a sample case study, these are the objectives and design targets agreed in scope — not measured business results.
PLACEHOLDER — Catalogue coverage
Target only
Placeholder. Replace with a measured, verifiable outcome or remove.
PLACEHOLDER — Publication gate
Human approval required
Placeholder. A constraint agreed during discovery.
PLACEHOLDER — Locales supported
Design target
Placeholder scope item, not a measured result.
Client feedback
In their words
Placeholder quote
This quote is sample content shown to demonstrate the layout. It is not attributable to a real person or organisation and will be replaced only with a genuine, approved testimonial.
“PLACEHOLDER — replace with a real, approved client quote. This text exists only to demonstrate the layout.”
Placeholder namePlaceholder role (illustrative)
Related services
The practices behind this work
AI & Automation
Applied AI engineering — assistants, agents, document workflows and automation — built on your data with evaluation and guardrails from day one.
Explore AI & AutomationWeb & Mobile
Web applications, mobile apps, commerce experiences and product design — built for performance, accessibility and conversion.
Explore Web & Mobile
More work
Other case studies
- HRTechSample content
An applicant tracking platform with explainable candidate matching
A recruitment platform where structured CV parsing and explainable matching cut screening time without removing recruiter judgement.
- Next.js
- TypeScript
- Node.js
- PostgreSQL
- FinTechSample content
Multi-tenant billing and entitlements for a growing SaaS platform
Retrofitting organisations, entitlements and usage-based billing into a product originally built for individual users.
- Next.js
- TypeScript
- Node.js
- PostgreSQL
- LogisticsSample content
Modernizing a logistics portal without pausing operations
Incremental migration of a business-critical customer portal and driver workflow using the strangler pattern, with no cutover event.
- Next.js
- TypeScript
- Node.js
- PostgreSQL
- HealthcareSample content
A patient portal that clinicians actually adopted
Placeholder case study. Appointments, results and secure messaging delivered against an immovable EHR, designed around clinician workload rather than around the feature list.
- Next.js
- TypeScript
- Node.js
- PostgreSQL
- E-commerceSample content
Replatforming a storefront without losing a peak season
Placeholder case study. A headless storefront migrated page type by page type, with checkout instrumented before anything was redesigned.
- Next.js
- TypeScript
- Shopify Hydrogen
- Algolia
- EdTechSample content
An assessment platform that holds up on results day
Placeholder case study. Timed assessment delivery engineered for concentrated load, with integrity controls and accessibility treated as requirements rather than settings.
- Next.js
- TypeScript
- Node.js
- PostgreSQL
- FinTechSample content
Cutting reconciliation break investigation from hours to minutes
Placeholder case study. Model-assisted investigation of reconciliation breaks, with every suggestion evidenced and every decision recorded for audit.
- Python
- TypeScript
- PostgreSQL
- Kafka
- HRTechSample content
Retiring a decade-old HR platform one capability at a time
Placeholder case study. A ten-year-old HR platform migrated capability by capability, with infrastructure brought under code before anything moved.
- TypeScript
- Node.js
- Next.js
- PostgreSQL
Similar situation?
Start a project like this one
Tell us where you are. We will tell you how we would approach it and what we would want to understand first.
Prefer email? contact@xalicon.co