SwitchPredict

SwitchPredict predicts which trains will need a crew recrew before they leave the terminal, allowing railroads to proactively add extra board crews and avoid service failures. Instead of relying on spreadsheets and tribal knowledge, operations and crew scheduling teams get a data-driven probability forecast for each train. The result: fewer unexpected recrews, less cascade delays, and more reliable service across your terminal network.

Built on an XGBoost model trained on 109K records across 33 features and 23 routes, the system delivers 96%+ accuracy with live predictions via a FastAPI service. A dashboard shows which trains are at risk so your dispatch center can act before problems happen.

Value Proposition

Predicts which trains will need a crew recrew (dog-catch) before they leave the terminal, so railroads can proactively add extra board crews and avoid service failures. Replaces tribal knowledge and spreadsheets with a data-driven, probability-based forecast per train.

Target Customer

Class I and shortline railroads — operations/dispatch centers and crew scheduling departments. US freight railroads lose millions annually to recrew-triggered delays and service failures.

Revenue Model

SaaS subscription based on terminal volume: $5k–$15k/mo per terminal tier, with enterprise pricing for railroad-wide deployment covering multiple terminals.

KPIs

Prediction accuracy – % of recrew events correctly forecasted vs. actuals

lead

Team Lead

Team management, task planning, delegation, and monitoring

Data Scientist

Builds and trains machine learning models for recrew prediction. Analyzes historical railroad crew data (on-duty times, HOS limits, recrew patterns, terminal congestion). Produces probability forecasts per train. Evaluates model accuracy. Can work with Python, pandas, scikit-learn, XGBoost, and time-series forecasting.

Full-stack Engineer

Builds web applications with React, TypeScript, Tailwind CSS, and backend APIs. Creates dashboard UIs, REST/GraphQL APIs, and integrates with data sources. Can deploy and manage infrastructure.

Connect to your terminal's crew and train operations data (on-duty times, HOS limits, terminal congestion patterns)
The system will begin generating live probability forecasts for each train departure
Monitor predictions in the included dashboard and integrate alerts into your dispatch workflow
No external MCPs required—all inference runs on your infrastructure via the FastAPI service on port 8000

Overview

Mission

Value Proposition

Target Customer

Revenue Model

KPIs

Progress

✅ Done

🔜 Next Steps

The team

Setup