Skip to experience
PSM / Chicago

Lead Data Scientist · Health Care Service Corporation

Pranav
Suresh Magadi.

I’m a Lead Data Scientist with 6+ years building scalable production AI systems for clinical and healthcare workflows—from foundational NLP to agentic systems. I combine hands-on engineering with technical leadership across architecture, delivery, and evaluation. At HCSC, I also chair the technical wing of enterprise AI governance, ensuring teams ship safe, fair, and high-quality AI while developing guidance and standards for emerging AI technologies.

Résumé
Pranav Suresh Magadi
Chicago, Illinois

How I work

  1. Deliberate system design

    I make deliberate architecture choices to balance accuracy, latency, and cost from the outset. I consider where to use AI, how to prepare context, and what can be reused or parallelized, then refine those choices through evaluation as the product scales.

  2. Business understanding and relationships

    I enjoy learning how a business works and building relationships with the people doing the work. At HCSC, that has developed into a deep understanding of payer workflows, including medical necessity reviews, utilization management, claims adjudication, DRG edits, care management, and underwriting. Combined with my earlier experience in insurance risk analytics, this helps me identify useful opportunities for AI and build solutions aligned with business needs.

  3. Lead development and delivery

    I have experience leading AI product development from architecture through production, working hands-on and collaborating with engineers on integration and deployment. I build robust evaluations and monitoring into these products to assess output quality and track reliability, latency, and cost in production.

Experience

Software development → Data science · 8+ years of experience

2022 — present

Health Care Service Corporation

Lead Data Scientistsince 2025

Previously Senior Data Scientist

Selected work

Selected work at HCSC

AI Risk Adjustment

One pipeline for chart screening and coder assistance

Context

Risk adjustment aligns payments to health insurers with the health risks of the members they cover, making complete, accurate diagnosis coding essential. Each diagnosis must be tied to the relevant visit’s date of service to support reimbursement. I built one AI pipeline to identify potentially risk-adjustable diagnoses and their supporting evidence in medical records, preserving that connection to the visit. It powers two use cases: Chart Screener prioritizes records with potentially missed diagnoses for a second coding review; Coder Assist gives internal coders searchable records, highlighted evidence, and structured summaries to support coding and audits.

My contribution

I led the shared pipeline’s architecture and development across standard medical records and a specific vendor’s format with more forms, tables, and checkboxes—keeping each diagnosis connected to its visit, evidence, and source.

Outcome

Chart Screener’s vendor-validated findings contributed approximately $1M in realized incremental reimbursement for the 2025 benefit year.

How it works

One extraction pipeline. Two review workflows.

Select a stage to see what it does.

Screening path
Batch prioritization spreadsheet
Coder Assist path
Highlighted records + summaries

Coders review the outputs and make coding decisions.

Read the records

The same pipeline handles standard medical records and a specific vendor’s format built around forms, tables, and checkboxes.

HEDIS Diabetes Models

Evidence extraction for quality review

Context

HEDIS—the Healthcare Effectiveness Data and Information Set—is a standardized set of measures used to evaluate and compare the quality of care delivered to health plan members. Claims data alone may not capture all the clinical information needed to assess these measures, so reviewers examine medical records to find the supporting evidence. I developed models that surface relevant evidence for diabetes-related measures—including blood pressure (BP), A1C, and retinopathy exams—helping reviewers assess whether the documented care meets the measure criteria.

My contribution

I developed the diabetes-measure models while our team rewrote the training pipeline from PyTorch to PyTorch Lightning. I worked through issues in the rewritten pipeline and trained annotators as they adopted a new workflow to produce annotations for the training run.

Outcome

Helps close HEDIS gaps through more efficient medical-record review, with up to 30% time savings measured in a time study.

How it works

Distributed training across eight GPUs.

Select a stage to explore.

Annotate evidence

Annotators label clinical evidence for measures such as blood pressure, A1C, and retinopathy exams. I trained and supported annotators as they adopted the workflow used to create this training data.

Enterprise AI Governance Chair

Chair, Model Approval Committee at HCSC

Mandate

I chair HCSC’s Model Approval Committee, responsible for technical review of every AI use case developed across the enterprise. I lead a team of data scientist reviewers, helping them apply consistent technical judgment across a wide range of AI systems.

My contribution

A substantial part of my role is researching emerging approaches—including retrieval-augmented generation and agentic AI—and translating that research into documented best practices and standards. These help teams build high-quality AI products, evaluate AI outputs effectively, and assess and address fairness and bias concerns.

How I support teams

  • Work with data scientist reviewers to assess AI use cases for technical quality, evaluation rigor, and fairness.
  • Lead and upskill data scientist reviewers for generative and agentic AI.
  • Keep the committee running efficiently and scale review capacity as needed so governance doesn’t become a blocker to shipping AI products.
  • Research emerging AI methods and document best practices and standards.
  • Help teams apply evaluation guidance through office hours and direct collaboration.

Before HCSC

My background combines software engineering, foundational NLP for insurance risk analytics, and healthcare AI, with an MS in Management Information Systems from the University of Illinois Chicago.

  1. 2020–2022

    Verisk Analytics

    Data Scientist I

    Foundational NLP and predictive modeling for insurance risk analytics.

    • Built BERT classifiers for legal case documents, turning unstructured text into liability categories for risk-estimation and loss-distribution simulations.
    • Developed generalized linear models and machine-learning models for property and casualty loss-cost forecasting and pricing.
    • Built multi-year projections for Lloyd’s Realistic Disaster Scenarios to support loss forecasts and reserve capital planning.
  2. Jul–Dec 2019

    University of Chicago

    Data Science Intern · Center for Translational Data Science

    • Developed a Random Forest model to predict fire events from NASA GOES-R geospatial and satellite imagery data.
    • Applied class-imbalance mitigation when training the fire-prediction model.
  3. Jun–Dec 2019

    Onco-care Analytics LLC

    Data Scientist Intern

    • Built data pipelines and interactive analytics for oncology providers in value-based care.
    • Developed Monte Carlo cost simulations to support cost analysis and risk-sharing strategies.
  4. 2019

    University of Illinois Chicago

    MS, Management Information Systems

    Graduate study with data science coursework.

  5. 2016–2018

    Accenture

    Application Development Associate

    Software engineering.

  6. 2016

    BNM Institute of Technology

    BE, Computer Science Engineering

    India.

Pranav Suresh Magadi Chicago, Illinois