How Aislegleam works
From raw catalog data to personalized storefront in under 200ms. A walkthrough of the four-stage pipeline that runs per session.
Four stages from data to delivery
Data Ingestion
Catalog feed sync and clickstream event collection via lightweight tag.
Signal Processing
View, search, add-to-cart, and purchase events processed in real time per session.
Intent Modeling
Per-session shopper intent model built from session signals and catalog affinity.
Surface Delivery
Ranked product sets returned to homepage, PDP, search, and cart via API or widget.
Data ingestion and signal processing
Aislegleam ingests your product catalog through a scheduled feed sync, pulling SKU attributes, inventory status, pricing, and category hierarchy. Catalog changes and new arrivals are detected within hours of your next feed delivery.
Behavioral event collection uses a lightweight JavaScript tag that captures views, search queries, add-to-cart events, and purchases. Events stream into the processing layer in real time, with no batch processing delay.
Feed formats supported: JSON, CSV, and Shopify product API. Event collection: JavaScript tag with automatic deduplication and session stitching.
Every 4 hours by default. On-demand trigger available via API for flash sales and same-day arrivals.
Event latencyClickstream events are processed into the session model within 200ms of collection. No batch lag between shopper action and next recommendation update.
Data handlingSession-level signals only. No personally identifiable information is required or stored. Fully functional on anonymous sessions.
Intent modeling in session
Each session builds its own intent model in real time. The model combines three signal layers: session-specific behavior (what this visitor is doing right now), collaborative filtering (what sessions with similar patterns bought), and catalog affinity (which product attributes cluster with this session's engagement).
New sessions start from catalog-wide popularity and attribute signals. Intent sharpens within two to three page views. By the time a shopper reaches a product detail page or searches, the model has enough signal to re-rank meaningfully.
Model refresh cadence is per-event: every view, search, or add-to-cart triggers a re-rank of the shopper's surface. No polling or scheduled batch refresh.
Surface delivery endpoints
Ranked product sets are delivered to your storefront through three integration options. All three return the same ranked product data, just formatted for your rendering stack.
JavaScript Widget
Drop a script tag into your storefront. The widget renders product cards directly with your CSS. Zero build required. Best for teams without a custom front-end pipeline.
REST API
Call the recommendations API from your front end or server-side render. Returns a ranked array of product IDs and attribute data. Under 200ms p99. Supports server-side rendering for SEO-sensitive surfaces.
Shopify Liquid Tag
Native Liquid tag for Shopify storefronts. Renders the recommended product grid within your theme template. Works with Shopify's standard CDN and cache headers.