This is an experimental project based on my EDA hosted on Kaggle Product Recommendation EDA
The project is split into three services:
- Transactions service collects entries from the client including
customerId,productIdandquantity. The client uses a HTTP POST endpoint to create one or many transactions. Each transaction is stored in a Postgress database then sent to the message broker (RabbitMQ). - Generator services does the hard work by processing the transactions of a customer into a list of recommended products. As this process could take some time, it works asyncronously by consuming the message broker's transactions queue. Transactions could stack up in the queue and processed when the generator get around to them. Once processed the generator will send the recommendation to the broker's recommendations queue.
- Recommendations service picks up from the recommendations queue and stores the recommendations into a mongodb database. The purpose of using mongodb instead of postgress is because the recommendation is a somewhat instructured array of object that would not fit into a tabular model. Mongodb just stores the recommendations against a customerId. The recommendations API simply takes a customerID from the url parameter and returns the recommendations for that customer.
URL
transaction-service/transactions
Request Payload
{ "customerId": "CUS1", "productId": "PR1", "quantity": 4 }Response Payload
{ "ok": true }URL
recommendations-service/recommendation/:customerId
Response Payload
[
{ "P2": 5 },
{ "P5": 4 },
...
]The response payload is an array of objects where the key is the product name and the value is the rank. The rank reflects the recommendation confidence.
Use docker-compose up and use the respective localhost ports for each service:
- Transactions 3010
- Recommendations 3000