Kafeido

Kafeido is an AI platform, which is built on top of kubeflow and kubernetes, to provide real-time data streaming inferences with multiple machine learning models in low-energy consumption, low hardware spec requirement, and trustworthy ability.

Kafeido, a nicked name for kaleido, is a tunnel which wires datasources with machine learning models and support the following mapping:

• One-datasource-one-model
• One-datasource-multiple-models
• Multiple-datasources-one-model
• multiple-datasources-multiple-models

And we believe that this could serve the most applications right now we are facing.

Kafeido can be deployed on top of multikf (a command line library to provision kubernetes cluster on the single host machine) or be deployed with kubeadm which is kubernetes’s official deployment command line tool for kubernetes. Please contact us if you want to provision Kafeido on your own machines.

The minimum hardware requirements for server is 8 Core+ CPU, 18G+ system memory, 50G+ Disk spaces, and GPU accelerators. There are no hardware requirements for its client.

Usage

Create Project

create project --name $projectname \
--desc "project for demonstration"

Create Datasource

./cli create datasource --project_id=$project_id \
 --bucket_name=$project_bucket \
 --index_object="videos/<some-video-prefix>.mp4" \
 --duration_per_frame=1s \
 --fps=1 \
 --type=VIDEO

Create Model

./cli create inference --project_id=$project_id \
 --bucket_name=$project_bucket \
 --model_uri=kafeido:///$project_bucket/$object_path

Create Prediction

 ./cli create predict --project_id=$project_id \
 --inference_id=$inference_id \
 --query_file=example.wav --query_type=audio --query_lang=en