End-to-end Product Flow

Lumiata's product suite enables an end-to-end machine learning lifecycle journey for healthcare organizations. Our aim is to simplify each step in the lifecycle so that Data Scientists, Data Analysts, and Citizen Data Scientists can focus on building and deploying models rather than worrying about data prep, infrastructure, and deployment.

Our five-step approach gives users the autonomy to seamlessly navigate Lumiata's AI Platform without worrying about constant starts and stops.

These five steps are as follows:

Step 1: Data Ingestion & Preparation

Each customer engagement begins with the ingestion of data into the Lumiata AI Platform where each customer will be assigned a dedicated, single-tenancy environment. You will share your data dictionary so that our Data Engineering team can map your organization's data to Lumiata's Data Model, a longitudinal record of each person in your records called Person360: Longitudinal Person Records).

Lumiata's Data Model is extremely flexible, so if there are custom data elements that need to be added for your organization, please reach out to your Customer Success Manager for support.

Next, the raw data from your organization will be ingested via sFTP or a direct connector into your data warehouse. We can ingest data in the following ways:

  • Customer's sFTP
  • Lumiata's sFTP
  • Data Warehouse Connector
  • Cloud Storage Bucket (Google, AWS, Azure)
    **See details in the Data Upload section.

Once the raw data is ingested, our Data Engineering team will begin to evaluate it for data quality. Examples of this include:

  • Data Completeness
  • Data Sufficiency
  • Data Accuracy
  • Record and row counts
  • Claims Distributions
  • Syntax

A data quality report will then be made available for the end-user in the platform.

After the raw data quality assessment is complete, the data is enriched and cross-codes are applied, normalized, and transformed into Person360 records. A quality check of the mapping is also completed to ensure the correct mapping of raw data to Person360.

Step 2: Data Exploration

Once the data has been transformed into Person360 records, you'll be able to explore the data using the Healthcare Data Management module, which allows you to access the following:

  • Visuals for Exploratory Data Analysis (EDA)
  • Detailed Person360 Records
  • Query Functionality

For our advanced Data Scientists, you can also perform EDA in a pre-configured Jupyter Notebook as well as create your own EDAs. You can import any library you need to create custom visuals.

In order to maintain HIPAA compliance, we can not allow you to download data to your machines; however, data can be pushed to your sFTP client for offlineA access to information.

Step 3: Model Training

After performing your EDA, you're ready to build a model. Lumiata's AI Platform was built with the core tenant of "democratize AI for healthcare". Data Analysts and Citizen Data Scientists can train and deploy models for healthcare use cases by utilizing our user interface (UI). For advanced Data Scientists, we offer integration with JupyterHub for complete freedom to move around the product as needed. Any user is able to access the pre-configured notebooks available in JupyterHub as a starting point for EDA, feature engineering, and model training.

Lumiata has pre-engineered healthcare-specific features for model training. We're able to expand the feature space beyond a customer's in-house dataset with data enrichment, cross-coding, and disease tagging,

Data Analysts and Citizen Data Scientists are able to build ML models by using the UI. A pre-configured interface guides the user through the process of selecting data, selecting a learning problem, and configuring an experiment. Once an experiment has been configured, a user can create their first run of the experiment. Each run allows a user to select different features, time slices, and hyperparameters, and users can create additional runs to optimize for better performance.

Advanced Data Scientists may also opt for a notebook through Jupyter integration. They can begin with a pre-configured notebook created by Lumiata or they can start from scratch. Either way, they have full access to all of Lumiata's libraries plus the flexibility to download third-party packages.

Step 4: Model Deployment

Once a model has been trained, tested, and validated it is ready to published. A model can be published either from the UI or through the notebook. The model binary will be registered within the platform and is then ready to be served with new data. The published model can be located in the Spectrum AI Model Catalog. Lumiata also offers users the option to deploy models into 3rd-party applications using APIs.

Step 5: Prediction Results

Prediction results can be viewed in the Predictions Viewer portion of the platform. Prediction analysis is available through the UI with a summary of the prediction results along with detailed records. Output of the predictions can be pushed to your organization's sFTP site or integrated with a call to the API.


Did this page help you?