KAI Enablers
This project corresponds to AI Platform's Generative AI & Knowledge AI MLOps journey.
Project Goals
Kai-Enablers is an event driven application/model serving framework through which various Knowledge AI models and services are made available to Equinor users/applications in a scalable manner. This will facilitate reuse and integration of key software artifacts & language models.
Summary of Results
Kai Enablers' vision is to Empower employees to derive meaningful information and insights from Equinor’s vast amounts of unstructured data, in order to improve the quality of decision making, make recommendations and solve business challenges using that information.
Project Team
Jennifer Sampson (Data Scientist)
Peter Koczca (Data Scientist)
Terje Elde (Kubernetes Developer)
Prerit Shah (MLOps Architect) Matthew Li (MLOps Architect)
MLOps Challenges
Multiple Clients Sending Requests
Multiple software applications and users would be making requests to Kai-enablers framework for predictions as per their needs and usecase. It is necessary to keep track of all the requests and where it came from to process them efficiently.
Decouple Applications using Microservice Architecture for Re-Use
Each model needs to be its own microservice and needs to be independely deployed as an endpoint for it to be used by multiple clients.
Need to Autoscale Based on Traffic
Workers need to automatically scale up based on number of requests.
Heavy Compute Requirement
Needed GPU as compute for training or inference of certain models
MLOps Solutions
GPU Compute on Kubernetes Cluster
Team was able to make use of GPU compute on Kubernetes to leverage the scale and power of kubernetes to process the ML model training jobs.
Event Driven Architecture
Azzure Service Bus Queues were used to send requests and receive response over gRPC protocol.
KServe Serverless ML Inference Framework
KServe was used to deploy each model as an independent endpoint with built-in autoscaling capability.
GitHub Repos
https://github.com/equinor/equibus/tree/initial_implementation
https://github.com/equinor/kai-enablers-models-transformers/tree/dev
https://github.com/equinor/kai-enablers-https-ingress/tree/dev
https://github.com/equinor/kai-enablers-selenium-scraper/tree/dev
https://github.com/equinor/equibus-kserve/
Demos
https://frontend-kai-enablers-frontend-demo-dev.playground.radix.equinor.com/
https://frontend-kai-enablers-frontend-demo-dev.playground.radix.equinor.com/Entity_Extractor
