AI offering engines Review and analyze information in the knowledge database, deal with the implementation of design and display display. They represent the entire new world in which applications will be able to use AI innovations to increase operational effectiveness, as well as to solve significant service problems.
Perfect practices
I was dealing with Redis Labs clients to better understand my obstacles in taking artificial intelligence in production, as well as how to design their engines offering artificial intelligence. To help, we have created a list of the best techniques:
Fast portion to the end
If you support applications in real time, you need to make sure that adding AI's ability at a stack will certainly have a small or no effect on the performance of the application.
No downtime
Since each contract potentially includes some AI processing, you must maintain a regular standard SLA, preferably at least five nine (99.999%) for critical applications of missions, using proven mechanisms such as duplication, data perseverance, multi schedule zone, active active zone, active circulation, regular backups and automatic cluster copies.
Scalability
Numerous applications are built by customer activities to support peak instances, from Black Friday to a great game. You require versatility to scale or scale the engine offering AI on the basis of expected and current tons.
Help for many systems
Your AI engine must be able to operate deep models trained by the latest systems, such as Tensorflow or Pytorch. In addition, machine learning projects, such as random-forest, as well as a linear reference, still ensure good predictability for many instances using, and must also be maintained by the AI engine.
Easy to place completely new models
Most companies want alternatives to often update their versions according to market trends or manipulate completely new possibilities. The update of the version should be as transparent as feasible, and also cannot affect the efficiency of application.
Monitoring and re -training performance
Each person must know how a model has developed well, and can also tune it in accordance with how good it is in real life. Make sure that AI offers AI to test A/B engine to contrast the version compared to the default model. The system should also provide tools for the AI implementation ranking of your applications.
Free yourself all over the world
Most of the time it is best to develop, as well as learn clouds, and also be able to offer anywhere as an example: in the supplier cloud, in many clouds, local clouds or on the edge. The AI engine should be an agnostic platform based on innovations related to open resources, and have a well -known version of the version that can work on processors, advanced graphic processors, engines, and even the Raspberry Pi device.
