Customer Data Relationships
Break the data custody logjam and deliver for more customers faster.
The Data Access Problem
Data is the lifeblood of advanced machine learning providers, and it's becoming harder and harder to obtain.
- Data Security
Customers' most useful data is confidential, regulated, or otherwise sensitive. They are less and less willing to part with it.
- Data Gravity
Unstructured/real-time data can be very large. Moving it to a service provider for processing is costly or infeasible.
- Value Paradox
You need to be able to show value to get customers to share data, but how can you show value without already having models trained on their data?
DON'T FIGHT THE DATA
If you can't bring the data to the model, bring the model to the data.
Using a federated deployment architecture enables you to design data relationships that work for both you and your customers.
Customer friendly data relationships
Meet your customers where they are and avoid data custody battles.
Local Data Plane for Inference
Customers get your insights without their data ever leaving their premises.
Base Model Data Partnerships
Work together to craft safe data partitions to enable federated training for base model improvements.
Customer Data Access Control
By keeping data custody, customers can add, change, or revoke data access instantly.
Provider friendly business models
Keep all the advantages of SaaS business models while satisfying customer data constraints.
Usage-based Licensing
Traditional "on-prem" licensing models are blunt and don't align with MRR/usage focused billing models. proxiML enables SaaS business models with on-prem data security.
Easy Model Versioning
Update deployed models in an instant. Programmatically retrain customer fine-tuned models and redeploy across all customers.
Centralized Control Plane
Manage and support 1000s of separate deployments through a single pane of glass.
Instant Model Access Control
The customer never takes custody of your models. You always have control over when they're deployed.
Customer Managed Scale
Customers control the resources available in their deployment. If they want higher performance, they can add more resources. You don't lift a finger or pay a dime.
Maintaining customer deployments is hard to scale
SaaS architecture has been dominate for a reason. Managing and deploying applications to customer's infrastructure is tough.
One is hard, 1000s is prohibitive
Packaging your application as downloadable software is hard enough. Supporting 1000s of customers with heterogeneous infrastructure and limited engineering skills isn't worth the effort.
Customer IT Constraints
Even with a great on-boarding process, customer IT resources are stretched thin. If you require anything more than user credentials, you'll be waiting 18 months to get started.
ML Models are Different
"Releasing" a new model version is more complex than publishing a software version update. Models weights can be very large, and fine-tuned models need to be retrained on customer data before they are ready for inference.
Let proxiML make it simple
We provide the infrastructure automation, user experience, data management, and security necessary to implement a federated service architecture as easily as a centralized one.
