E-Commerce Product Discovery & Personalization
HeliosDB Lite powers product discovery for an online store: each product row carries an embedding of its name and description in a VECTOR column, an HNSW index finds the products closest in meaning to a shopper's search text, and ordinary predicates such as category and maximum price filter the same SQL query, so search and catalogue data stay in one database.
Executive Summary
HeliosDB Lite enables e-commerce platforms to deliver personalized product discovery combining search + recommendations + inventory in a single embedded database. This eliminates the need for separate recommendation engines (collaborative filtering systems) and search platforms. Key value propositions:
- Personalized product rankings based on semantic similarity + purchase history
- Real-time inventory-aware recommendations (no stale product listings)
- JSONB product attributes enabling flexible catalog management
Problem Being Solved
The E-Commerce Personalization Dilemma
Option B: Collaborative Filtering (ML Recommendation Engine)
- Pro: Personalized recommendations
- Pro: Discovers products users don't search for
- Con: Slow (batch processing) - recommendations hours old
- Con: Cold start problem (new products/users have no data)
- Con: Complex infrastructure (ML pipelines, data scientists)
E-Commerce Pain Points
Operational Burden:
- Data scientists maintaining ML models
- ETL pipelines syncing inventory to recommendation engine
- Cold start problem for new products (no recommendations for 2 weeks)
- Inventory mismatches (product data drifts between systems)
HeliosDB Lite Solution Architecture
Unified Product Discovery Platform
┌──────────────────────────────────────────────────────┐
│ E-Commerce Platform (Backend) │
├──────────────────────────────────────────────────────┤
│ │
│ HeliosDB-Lite (Embedded) │
│ ┌────────────────────────────────────────────────┐ │
│ │ Products Table │ │
│ │ ├─ product_id (PRIMARY KEY) │ │
│ │ ├─ name, description, price (TEXT) │ │
│ │ ├─ category_id, subcategory (VARCHAR) │ │
│ │ ├─ embedding (VECTOR) [semantic] │ │
│ │ ├─ attributes (JSONB) [color, size, material] │ │
│ │ ├─ inventory (INT) │ │
│ │ ├─ margin, profit (FLOAT) │ │
│ │ └─ created_at, updated_at (TIMESTAMP) │ │
│ ├────────────────────────────────────────────────┤ │
│ │ Customer Preferences Table │ │
│ │ ├─ customer_id (PRIMARY KEY) │ │
│ │ ├─ purchase_history (JSON array) │ │
│ │ ├─ browsing_history (JSON array) │ │
│ │ ├─ preferences (JSONB) [colors, styles, etc] │ │
│ │ ├─ embedding (VECTOR) [customer taste profile] │ │
│ │ └─ updated_at (TIMESTAMP) │ │
│ ├────────────────────────────────────────────────┤ │
│ │ Indices │ │
│ │ ├─ Vector HNSW (product similarity) │ │
│ │ ├─ Composite (category + price) │ │
│ │ ├─ Inventory (in-stock filtering) │ │
│ │ └─ Temporal (trending, seasonal) │ │
│ ├────────────────────────────────────────────────┤ │
│ │ Real-Time Personalization Engine │ │
│ │ ├─ Customer taste profile matching │ │
│ │ ├─ Purchase history analysis │ │
│ │ ├─ Browsing behavior patterns │ │
│ │ └─ Collaborative signals (similar customers) │ │
│ └────────────────────────────────────────────────┘ │
│ │
│ Recommendation Pipeline (SQL-based) │
│ ├─ Personal recommendations (top 20) │
│ ├─ Similar products (for product pages) │
│ ├─ Trending products (dynamic, real-time) │
│ └─ Cross-sell/up-sell (based on cart) │
│ │
└──────────────────────────────────────────────────────┘
↓ (REST API / WebSocket for real-time updates)
┌──────────────────────────────────────────────────────┐
│ E-Commerce Frontend (Web/Mobile) │
├──────────────────────────────────────────────────────┤
│ ├─ Product search │
│ ├─ Personalized homepage recommendations │
│ ├─ Product detail page (similar items) │
│ ├─ Shopping cart (up-sell suggestions) │
│ ├─ Checkout (cross-sell offers) │
│ └─ Order confirmation (personalized thank you) │
└──────────────────────────────────────────────────────┘Real-Time Recommendation Queries
-- Personalized recommendations for customer
SELECT
p.product_id, p.name, p.price, p.category_id,
-- Semantic similarity to customer preferences
(1 - (p.embedding <-> cp.embedding)) * 0.5 as preference_score,
-- Purchase history relevance
CASE WHEN p.category_id IN (SELECT category_id FROM customer_purchases)
THEN 0.3 ELSE 0 END as category_score,
-- Popularity boost
(CASE WHEN p.inventory > 100 THEN 0.1 ELSE 0 END) as availability_score,
-- Margin preference (recommend high-margin items)
(p.margin / 100) * 0.1 as margin_score,
-- Final ranking score
(preference_score + category_score + availability_score + margin_score) as rank_score
FROM products p
CROSS JOIN customer_preferences cp
WHERE cp.customer_id = $1
AND p.inventory > 0 -- Only in-stock
AND p.price BETWEEN $2 AND $3 -- Price range preference
AND p.product_id NOT IN (SELECT product_id FROM customer_purchases) -- Don't recommend already owned
ORDER BY rank_score DESC
LIMIT 20;Conclusion
HeliosDB Lite transforms e-commerce from "search-centric" to "discovery-centric" by embedding personalized product recommendations directly into the application.
Performance depends on workload and hardware. Published figures are measured on representative fixtures; see the benchmarks page and reproduce them on your own workload.