Simple and robust ML models deployment
Automated versioning
Easy models and versions management
Score the model from your app or microservice via REST, gRPC or Kafka stream API.
A/B and Canary testing on production traffic.
Hot-wing bumpless model replacement in production pipeline
Hydrosphere.io Platform for AI/ML Operations Automation
1. AI / ML Operations Automation
SUSTAINABLE. OBSERVABLE. SCALABLE.
2. About Us
Hydrosphere.io emerged as autonomous
project in 2016, congregating experience
and solutions accumulated delivering
professional services in AI/ML sphere on
the USA market.
Team expertise includes Software
Development, DevOps and Data Science.
Parent Company: Founded: 2011 / Revenues: $20M (2017)
provectus.com
4. Hydropshere.io - Enterprise Grade Model Management Platform, Open Source
Simple and robust deployment
Automated versioning
Easy models and versions management
Score the model from your app or
microservice via REST, gRPC or Kafka stream
API.
A/B and Canary testing on production traffic.
Hot-wing bumpless model replacement in
production pipeline.
Training framework does not matter: we
serve Tensorflow, Scikit-learn, Keras, PyTorch
and other runtimes, including custom Python
models.
Infrastructure agnostic - if your premises or
cloud service support Docker - you can run
our platform.
Kubernetes is supported (optional).
Production traffic and model performance
monitoring. Training-production data skew,
outliers, concept drifts and new concepts
detection.
Advanced statistical methods and deep
learning methods to monitor and sustain
inference quality.
Automated profiling and subsampling to
efficiently retrain ML models.
★ GitHub: https://github.com/Hydrospheredata/hydro-serving
Description: https://hydrosphere.io/ml-lambda/
Documentation: https://hydrosphere.io/serving-docs/
5. A Way of
Machine
Learning
Build a Model
Set up an Infrastructure
Train,
Tune
Validate
Architec-
ture
Engine-
ering
Data flows
plumbing
Running,
Scaling
Sustain accuracy & speed
Observe
perfor-
mance
Detect
anomalies
Trigger
retraining
6. Struggle for ML Production
1. Data Science field:
● Analyse and prepare data
● Train and tune models
● Watch AI/ML production quality
● Build subsamples from production data
● Trigger model retraining process
2. DevOps field:
● Build and commission infrastructure
● Perform dataflows plumbing
● Deploy and re-deploy models
7. Struggle for ML Production
1. Data Science field:
● Analyse and prepare data
● Train and tune models
● Watch AI/ML production quality
● Build subsamples from production data
● Trigger model retraining process
2. DevOps field:
● Build and commission infrastructure
● Perform dataflows plumbing
● Deploy and re-deploy models
10. ML Function As a Service on Premises or in Cloud
★ GitHub: https://github.com/Hydrospheredata/hydro-serving
Description: https://hydrosphere.io/ml-lambda/
Documentation: https://hydrosphere.io/serving-docs/
11. Sonar: Observable ML Production
It is a closed solution for Enterprise Delivery
Along with statistical ones, ML-algorithms are comprehended to detect
anomalies: GAN, MADE, Auto-Encoders.
Description: https://hydrosphere.io/sonar/
13. AdTech Company Case: Results Delivered
● Machine Learning operations got scaled from 2 models to 200+ models in
production
● Stabilised and solidified Machine Learning pipelines gave $20M of annual
savings.
● ML Team productivity doubled, estimated ROI increase is $1M per year.
● Data science production iterations went seamless saving min. 2 weeks of time
per release.
● The demand for DevOps people presence in release chain was eliminated
completely delivering a solid improvement to costs and ROI.
● A month of man-hours for product management and a 3 months for QA are
saved per release.
● Apache Spark jobs completion rate reached 99%.
● Cluster throughput increased 10 times saving $100K monthly.
● Facilitating over 10 products, implementation of the hyrosphere.io platform into
AI/ML operations created a new revenue stream of $10M annually.