Skip to content
Research & DataAward Winner

RouteRx (Careline) — Bi-Directional Transit Hub

Role
Lead Data Scientist & Spatial Systems Engineer
Dates
2026
Stack
RStudio · Python · Tigercensus · OpenStreetMap · GTFS · OSRM · React
RouteRx (Careline) — Bi-Directional Transit Hub

01.Executive Summary

RouteRx (Careline) is a spatial logistics engine designed for Stormont Vail Health in Topeka, KS. By unifying preventative patient pick-up routes with discharge return trips on a single bi-directional shuttle fleet, it eliminates healthcare transit deserts while clearing acute hospital bed bottlenecks.

02.Problem & Constraints

Patients in transit-isolated census blocks miss preventative appointments and escalate into preventable ER emergencies. Simultaneously, medically discharged inpatients are trapped occupying acute hospital beds because public transit stops are inaccessible. In Topeka, 43 of 141 census blocks had zero bus stops within an 800m walking radius.

03.System Architecture

Engineered a spatial analytics workflow combining Tigercensus demographic data, OpenStreetMap road networks, and GTFS transit schedules in RStudio and Python. Formulated a Transit Vulnerability Index (TVI) and applied DBSCAN density clustering to identify patient pickup zones, routing shuttles through Open Source Routing Machine (OSRM) turn-by-turn geometry.

Interactive architecture laboratory

RouteRx (Careline) Spatial Data Pipeline (C4 Model)

Multi-source geospatial data fusion pipeline driving bi-directional routing manifests.

C4 Level 2/3 Component Model
externalTigercensus / OSM / GTFS

External Geodata

Pulls census demographics, bus stop coordinates, and road network edges

workerRStudio / Tigercensus

TVI Calculator

Calculates Transit Vulnerability Index for all 141 Shawnee County blocks

workerPython / Scikit-Learn

DBSCAN Spatial Clusterer

Groups high-vulnerability residential blocks into optimal pickup nodes

apiOpen Source Routing Engine

OSRM Routing Machine

Generates turn-by-turn driving paths and arrival ETA manifests

clientReact / Mapbox GL

Hospital Dispatch UI

Visualizes transit deserts and generates daily driver passenger manifests

Inter-Component Protocol Boundaries

data-sources→tvi-engine
Batch ETL
tvi-engine→dbscan-clusterer
GeoDataframe
dbscan-clusterer→osrm-router
REST GeoJSON
osrm-router→dispatch-dashboard
JSON API

04.Engineering Decisions & Tradeoffs

Aggregated million-coordinate spatial geometries into regional bounding boxes and discrete shuttle stop clusters, trading micro-level coordinate fidelity for real-time routing calculation speeds.

Decision 1: Bi-Directional Fleet Utilization vs Separate Fleets

Option chosen

Unified shuttle manifest pairing morning preventative pickups with afternoon discharge returns.

Alternative rejected

Commissioning two dedicated independent shuttle fleets.

Cost accepted

More complex routing optimization and dynamic manifest rescheduling.

Measured impact

Cut simulated transportation costs by ~40% while freeing acute hospital inpatient capacity.

Decision 2: DBSCAN Spatial Clustering vs Fixed Administrative Zones

Option chosen

Density-based spatial clustering of vulnerable patient centroids.

Alternative rejected

Grouping by standard ZIP code or municipal neighborhood boundaries.

Cost accepted

Clusters do not strictly align with existing municipal district boundaries.

Measured impact

Revealed contiguous transit-starved micro-clusters that cross ZIP code boundaries.

05.Evaluation & Verified Metrics

Key performance indicators and latency profiles benchmarked under explicit workload and hardware constraints.

Award

1st Place

Best Use of External Data (ASA DataFest 2026)

Measured on: The Ohio State University, College of Arts & Sciences.

Transit Deserts

43 of 141 Blocks

Blocks with 0 bus stops within 800m

Measured on: RStudio, OSRM routing engine, Python spatial libraries.

Routing Engine

OSRM Optimized

Bi-directional patient pickup & discharge routing

Fleet Optimization

~40% Cost Cut

Simulated savings from dual-use shuttle fleet

06.Personal Contribution

Co-led spatial modeling with Samay Patel, Harsha Kalivarapu, and Sudhish Gopalakrishnan. Authored the Transit Vulnerability Index formulas, performed GTFS transit accessibility buffer analysis, and structured the React dispatch dashboard.

07.Limitations & Next Steps

Known limitations:

Assumes normal road weather conditions and standard walking capabilities; does not account for temporary winter snow closures or steep topographical barriers.

Next engineering steps:

Integrating real-time vehicle GPS tracking and wheelchair-accessibility constraint filters into the active OSRM dispatcher.