Schedule a Call
  • HEALTHTECH
  • AI & DATA ENGINEERING

Accelerating Detection Speed by 50% Through Custom Data Pipelines and ML Models

How Bluelight built a scalable, explainable AI pipeline that helps Vasc‑Alert catch dialysis complications early — transforming patient care with real‑time insight.

Explore Case

The Challenge

Vasc‑Alert had vast amounts of data from millions of dialysis sessions, but data inconsistencies and manual entry errors hindered analysis.

Existing processes could identify only up to 30% of adverse cases, leaving too many complications undetected until it was too late.

File-based data silos made integration and processing cumbersome, slowing down every downstream analysis.

Main Service

Software Development

Technologies Used

  • Python
  • GCP
  • Vertex AI
  • DataFlow
  • BigQuery
  • Power BI
  • Tableau
  • Kubeflow

Our Process

To unlock the potential of Vasc‑Alert’s data, Bluelight designed an end-to-end pipeline spanning ingestion, analytics, modeling, and production deployment.

  1. Data Pipeline Implementation

    Built DataFlow pipelines to process and clean data from both automated and manual sources.

  2. Advanced Analytics & Reporting

    Centralized data in BigQuery for fast, real-time analytics with seamless integration into Power BI and Tableau.

  3. AI Model Development

    Used boosted tree models to accurately predict adverse incidents, improving early detection by 50%.

  4. Productionizing the Model

    Built a robust training pipeline with Kubeflow Pipelines for containerized, automated retraining and monitoring.

Main Goal

Develop a reliable, explainable AI model capable of detecting dialysis complications early — despite inconsistent, siloed, and partly manual data sources.

Chart: scores and interventions over time for one patient
Custom visualization of patient historical data with classic and upgraded scoring for undesired events. Vertical red lines show recorded medical interventions.
Chart: classic score versus machine learning score for interventions within 60 days
Classic score (legacy) and machine learning score (Bluelight model) for interventions within 60 days. Our model was not only much more accurate, but also easier to understand thanks to its linearity and close-to-zero y-intercept.

Our Impact:

Increased early detection
50%
of diagnosed patients require dialysis
97%
of adverse cases caught by legacy process
30%

Revolutionizing dialysis patient monitoring

Bluelight’s solution transformed Vasc-Alert’s data challenges into opportunities by combining pioneering Data Engineering and Machine Learning to equip medical professionals with the tools they need to make proactive, data-driven decisions. By advancing dialysis complication prediction with a game-changing AI model, we are helping shape a healthier future for kidney disease patients across the US.

  • Scalable, automated data pipeline delivering consistent data quality and integration across sources.
  • Intuitive visualizations and automated reports gave medical staff deeper, more proactive insight into patient health.
  • Full automation across disaster recovery, updates, patching, and infrastructure-as-code implementation.
  • The model’s linear, close-to-zero-intercept scoring made it significantly easier for clinicians to interpret than the legacy system.

Explore more of our work

See All Cases

Do you want to be our next success story?

Tell us the skills you need and we'll find the best developer for your needs in days, not weeks.

We’ll assemble a senior team in days, not weeks, with a 2-week risk-free trial. Not the right fit? You don’t pay.