QUALIGEEKS (OPC) PVT. LTD. · HEALTHCARE TECHNOLOGY · 2022–2024
Making healthcare scheduling
more responsive.
Work across predictive ML, event processing, and cloud applications supporting day-to-day scheduling decisions.
Better signals.
Faster response.
01 / THE CONTEXT
Scheduling depends on timely decisions.
Healthcare teams need to spot appointments at risk of being missed, respond to cancellations, and keep availability up to date. At Qualigeeks I worked across three parts of a broader scheduling platform: prediction, event processing, and the application services behind those workflows. These were related workstreams, not one continuous ML pipeline.
02 / THE WORK
Three areas of the scheduling platform.
Appointment no-show prediction
I developed appointment features in Databricks and worked on XGBoost training and evaluation with SageMaker. The team used temporal validation and nightly batch inference to score more than 50,000 upcoming appointments, with a separate path for real-time rescoring.
Cancellation-event processing
I worked on Amazon MSK and Spark Structured Streaming pipelines for cancellation events and slot-release updates. Idempotent writes helped manage duplicate processing. The cancellation-to-slot-release processing stage improved from minutes to sub-second.
Application and cloud engineering
I contributed to Python API and PostgreSQL scheduling workflows and helped operate AWS EKS environments with Terraform and CI/CD. Those environments supported roughly 25 microservices and more than 300 pods across the broader platform.
03 / RESULTS
Better prediction, faster processing.
In an offline temporal evaluation, no-show recall increased from about 62% to 74%. The nightly batch-scoring process assessed more than 50,000 upcoming appointments. The event-streaming improvements were measured on cancellation-to-slot-release processing, not the complete user-facing scheduling journey.
Recall is a held-out evaluation result, not a guarantee for every future appointment. The streaming improvement refers to cancellation-to-slot-release processing rather than the complete booking experience.
Generalized case study for portfolio use. Project details and metrics should be shared publicly only when permitted by the relevant employer or client.