Integrated production AI
Integrated AI models, image processing and LLM services into dependable production stages.
Autoscrapper
Catalogue growth for Verifyt
Turning fashion data into richer shopping experiences.
Autoscrapper supplies Verifyt, an AI-powered fashion social app, with structured products and imagery from 100+ top US fashion brands. The catalogue powers discovery, personal wardrobes, social content and affiliate shopping, one of the app’s main revenue sources.
1M+product records and images processed
100+top US fashion brands in the catalogue
Business-criticalinfrastructure supporting a main revenue source
Growing the platform
Better catalogue coverage opened new opportunities
Richer, current products expanded Verifyt’s consumer and influencer reach, leading to more brand relationships and affiliate access.
Different product pages, uneven affiliate-feed rows, mixed image types and inconsistent variants enter Autoscrapper. They leave as one validated product family with selected imagery, grouped sizes and colours, a reliable category and a current shopping destination for Verifyt.
A richer catalogue makes Verifyt's shopping and wardrobe experiences more useful. This expands consumer and influencer reach, supports more brand relationships and opens affiliate access through CJ, Rakuten, Skimlinks, Impact and Awin.
Applied AI
Several AI capabilities working as one production system
Extraction, language and image processing turn varied source data into products ready for Verifyt.
The production sequence extracts names, variants, materials, prices and availability; enriches descriptions and attributes with an LLM; detects the strongest product, model and flat-lay images; prepares background-removed assets; classifies categories and attributes; then publishes only records that pass quality checks. When no flat-lay image is available, classification uses the strongest display image.
Integrated specialised AI models into Autoscrapper and adapted their interfaces for dependable processing. Built Docker images for GPU workloads with CUDA requirements.
Operational visibility
Control complex product processing from one interface
Operators select products and processing stages, then monitor current counts from one interface.
The operator interface shows one running pipeline, a five-stage rail with Image analysis selected, current and optional stop-at stages, base and run-specific product selection rules, exclusions and fresh processing counts.
Faster brand onboarding
Using AI tools to reduce repetitive onboarding work
Operational MCP tools support onboarding, status checks and investigations. Fixed checks evaluate AI proposals before human review.
The operational workspace shows a product-page sample with detected fields, an AI-generated extraction-configuration proposal, fixed checks against development, held-out and risk samples, and a final state requiring human review. The same MCP-powered tools support onboarding, status checks and investigations.
Technical ownership
Connecting data, AI and production operations
Led the Autoscrapper team, delegating well-scoped implementation work while remaining hands-on with architectural decisions and major system changes.
Catalogue architecture
Connected varied sources to Verifyt's structured product catalogue.
GPU-ready operations
Built operator controls and GPU-ready deployment for production workloads.
AI-assisted developer tooling
Operationalised MCP tools for onboarding, status checks and investigations.