First, solve pure Bellman-Ford problems: detect cycles, find shortest paths with negative edges. Learn the basic algorithm before adding constraints. Then, solve constrained problems: limited stops, reachability conditions. These require modifying the standard algorithm while keeping the core logic.
Finally, apply to real-world scenarios: arbitrage, maximum paths, network routing. Each problem type reinforces different aspects of the algorithm. The more contexts you see, the better you understand when and how to apply it.