SakeWiz : Phase 2 Architecture
In the previous post, I shared the initial architecture of the SakeWiz platform. During the last 11 months, we have slowly implemented new features including a new “News Feed algorithm/component”, “Facebook login” feature, new “GPU powered label recognition system” ..etc
News Feed Algorithm
New “News Feed Feature” now shows activities from the followers as well. In order to implement this, all user events are captured and saved in an Event Log (similar to EventStore pattern https://microservices.io/patterns/data/event-sourcing.html)
GPU powered label recognition
The biggest change to the system was the new GPU powered label recognition component. Previous design used to run on a cluster of EC2 nodes (10 nodes) running a parallelized algorithm. New GPU powered algorithm running on a single NVIDIA Tesla K80 GPU. This increased not only the label recognition accuracy but also reduced the processing time. Apart from this, the AWS infrastructure cost was reduced by 40%.
The new algorithm was rewritten in C++ using OpenCV along with NVIDIA CUDA and rely on Pistache framework for HTTP/RPC. Amazon C++ SDK is used to communicate with S3 to retrieve product labels.
Facebook login
iOS and Android, Apps now support Facebook login for easy sign up/login.
Originally published at http://sakewiz.blogspot.com on November 9, 2018.
