This paper presents a reproducible Machine Learning (ML) pipeline designed for short-term cryptocurrency price prediction using multimodal data sources. The system integrates minute-level OHLCV data from Kraken with pre-scored Reddit sentiment, both resampled to hourly resolution and aligned to create a six-hour-ahead binary classification task. Twelve lagged hours of sentiment and price signals are appended to construct the feature set. Four classifiers–Support Vector Machine (SVM), Random Forest (RF), Extreme Gradient Boosting (XGB), and a majority-vote ensemble–are trained and evaluated using a walk-forward validation scheme. The results in the final test window show that the SVM produces the highest overall accuracy (52.05%), while the ensemble model achieves the best F1 score and recall (86.36%) in the minority “rise” class. Feature importance analysis indicates that historical price data dominate sentiment inputs, although all models benefit modestly from the added social signal. Although precision remains modest, high recall on upward trends highlights practical value for momentum-sensitive and risk-aware trading scenarios. Beyond modeling, this paper serves as a pedagogical tool in a senior software engineering course, emphasizing reproducibility, pipeline modularity, latency profiling, and data-centric system design. The fully automated Python framework, from ZIP extraction to model benchmarking, is openly available to support future enhancements, instructional use, and community-driven research.