E-commerce platforms rely on personalised product recommendations to increase sales, improve customer engagement, and deliver a better user experience. Modern AI models analyse customer behaviour, purchase patterns, search terms, price sensitivity, and market trends to generate accurate recommendations in real time.
This article explains how AI-based recommendation systems work, the types of models used, and how you can implement a scalable recommendation engine for an Angular + .NET + SQL Server–based e-commerce platform.
Why Recommendation Systems Matter
Increase product visibility
Boost cross-selling and up-selling
Reduce search friction for customers
Improve conversion rates
Increase customer retention
Deliver personalised shopping experience
A smart recommendation engine can predict and present items the user is likely to purchase next.
Types of Recommendation Systems
1. Content-Based Filtering
Recommends products similar to items viewed or purchased by the user.
Example: “Because you viewed Samsung mobile cases…”
2. Collaborative Filtering
Recommends based on behaviour of similar users.
Example: “Users who bought this also bought…”
3. Hybrid Models
Combines both approaches for more accurate recommendations.
4. Deep Learning Models
Neural networks predict what a user may want next based on user behaviour, embeddings, and sequence modelling.
High-Level Workflow of an AI Recommendation System
User browses or purchases products
User behaviour is captured
Data is stored in SQL Server or a data warehouse
AI model processes user activity
Recommendations are generated
API sends recommendations to Angular front-end
User sees personalised suggestions
Flowchart: Recommendation System Process
+------------------------------+
| User Activity |
| (Search, View, Purchase) |
+--------------+---------------+
|
v
+--------------+---------------+
| Capture behaviour & metadata |
+--------------+---------------+
|
v
+--------------+---------------+
| Store in SQL / Data Layer |
+--------------+---------------+
|
v
+--------------+---------------+
| AI Model Processes Data |
+--------------+---------------+
|
v
+--------------+---------------+
| Generate Recommendations |
+--------------+---------------+
|
v
+--------------+---------------+
| API sends results to Angular |
+--------------+---------------+
|
v
+--------------+---------------+
| Display personalised items |
+------------------------------+
Architecture Diagram (Visio Style)
+-----------------------------+
| Angular Front-End |
| (Product Listing, Home) |
+--------------+--------------+
|
| API Calls
v
+----------------+----------------+
| ASP.NET Core API |
| Recommendation Controller |
+-------+---------------+---------+
| |
Data Access Model Service
| |
v v
+-------------+-----+ +---------------------+
| SQL Server DB | | AI Recommendation |
| (Users, Orders, |<---->| Engine (Python/.NET)|
| Products, Views) | | ML Model Inference |
+-------------+-----+ +---------------------+
|
v
+---------------------+
| Model Training Jobs |
| Azure ML / Python |
+---------------------+
ER Diagram (Recommendation Metadata)
+------------------+ +-----------------------+
| Users | 1---* | UserActivity |
+------------------+ +-----------------------+
| UserID (PK) | | ActivityID (PK) |
| Name | | UserID (FK) |
| Email | | ProductID (FK) |
+------------------+ | ActivityType |
| Timestamp |
+-----------------------+
+------------------------+
| Products |
+------------------------+
| ProductID (PK) |
| Name |
| Category |
| Price |
+------------------------+
+------------------------+
| Recommendations |
+------------------------+
| RecID (PK) |
| UserID (FK) |
| ProductID (FK) |
| Score |
| RecommendedOn |
+------------------------+
Sequence Diagram: Generating Recommendations
Customer Angular App .NET API AI Model Engine SQL Server
| | | | |
|---Visit----->| | | |
| |---API Req-->| | |
| | |---Fetch---->| |
| | | Data |---Query------>|
| | | |<--Data--------|
| | |---Send Data--------------->|
| | | | Predict |
| | |<--Recommendations-----------|
|<--Display Recom----------- | | |
| | | | |
Building an AI-Powered Recommendation Engine
Step 1: Collect User Behaviour
Track:
Viewed products
Search queries
Cart additions
Purchases
Time spent on pages
This data goes into SQL Server.
Step 2: Prepare Training Dataset
Use Python or Azure ML to prepare features:
Step 3: Train AI Models
Possible models:
Matrix Factorisation (ALS)
Neural Collaborative Filtering
Deep Learning with embeddings
Sequence models (RNN, LSTM)
Transformer-based recommenders
OpenAI embedding models (for similarity-based recommendations)
Step 4: Deploy Inference Model
The trained model can run in:
Python FastAPI microservice
.NET Web API using ML.NET
Azure ML Managed Endpoint
Step 5: Integrate with Angular
Angular will call:
/api/recommendations/user/{userId}
The API returns a sorted recommendation list.
Example JSON
{
"userId": 123,
"products": [
{ "productId": 11, "score": 0.92 },
{ "productId": 47, "score": 0.89 }
]
}
Using OpenAI Embeddings for Recommendations
OpenAI embeddings improve:
Similarity detection
Context understanding
Product matching
Example workflow:
Generate embedding for each product description
Generate embedding from user activity
Use vector similarity to find matching products
This gives highly relevant recommendations.
Best Practices
Use hybrid recommendation models for accuracy
Refresh recommendations periodically
Cache results for performance
Avoid over-recommending same items
Track user feedback to improve models
Log model behaviour for audits
Run A/B tests for UI placement
Conclusion
AI-powered recommendation systems are essential for modern e-commerce platforms. By combining behavioural data, deep learning models, and tools like OpenAI embeddings, businesses can deliver personalised experiences that increase sales and customer satisfaction.
Using an Angular front-end, a .NET API backend, and SQL Server as the data layer, developers can build scalable, intelligent recommendation engines suitable for high-traffic platforms.