TRANSIT2025
In production, $8/monthAutomated GTFS Realtime validation for St. Lawrence County Transit on a single $8/month EC2 instance. Terraform-provisioned, Docker-packaged, scheduled to service hours, with SNS alerting on critical feed errors and S3 report history.
GTFS-RTAWSTerraform+3
GEOSPATIAL2024
Delivered to 3 countiesA2A Corridor Roadkill Analysis
Identified wildlife–vehicle collision hotspots using kernel-density and penalized regression. Analyzed 1,000+ collision records with traffic, land cover, and environmental data. Impact: Planner-ready mitigation layers delivered to 3 counties.
ArcGIS ProGeoPandasscikit-learn+2
Client work — code not public
TRANSIT2024
Deployed for St. Lawrence TransitCost-Efficient GTFS-RT Pipeline
Deployed Lambda+S3 GTFS-RT pipeline ($4/month) with 30-day history for service and on-time performance analysis. Unified GTFS static/RT/Flex with automated QA. Architecture presented at MobilityData Conference 2024.
AWS LambdaS3GTFS+3
Client work — code not public
TRANSIT2024
Trip-level demand patternsExplored large-scale NYC taxi trip records with datetime decomposition and spatial-temporal visualization to surface demand patterns across time of day and pickup zone.
PythonBig DataMobility Data+1
MACHINE LEARNING2024
10M+ records analyzedAnalyzed 10M+ environmental audit records across US/EU/CA/IN and built Random Forest predictors for organizational emissions risk to inform carbon policy interventions.
Pythonscikit-learnML+2
TRANSIT2024
Interactive visualizationGTFS Transit Analysis StoryMap
Created interactive ArcGIS StoryMap showcasing transit network analysis and service patterns. Visualized route performance, ridership trends, and accessibility metrics.
ArcGIS OnlineStoryMapGTFS+2
Client work — code not public
TRANSIT2024
Deployed for VTCMicrotransit Zone Design
Designed zone-based microtransit service areas using demand analysis, travel-time modeling, and service constraints. Optimized zone boundaries for efficiency.
ArcGIS ProPythonNetwork Analysis+2
Client work — code not public
MACHINE LEARNING2023
88.3% accuracyCatBoost model achieved 88.3% accuracy predicting cancellations. Used feature importance and partial dependence to explain impact of lead time, ADR, and past cancellations.
PythonCatBoostML+1
GEOSPATIAL2023
Top 5% risk segments identifiedUnified crashes, traffic counts, pavement ratings, and environmental data into a single geodatabase. Computed segment-level features and a composite risk score to prioritize the top 5% of segments for intervention.
ArcGIS ProPythonPostgreSQL+2
MACHINE LEARNING2022
1M tweets processedClassified toxic content at scale using classical NLP (TF–IDF + linear models) and deep learning. Preprocessed 1M tweets with class imbalance handling and confusion-matrix analysis.
PythonNLPTensorFlow+2