While a Trie is the "correct" datastructure with appropriate storage techniques, you have to be careful when your hands are tied regarding what datastructure yo

While a Trie is the "correct" datastructure with appropriate storage techniques, you have to be careful when your hands are tied regarding what datastructure you build it on top of. When I benchmarked a Trie implementation using dicts of dicts in Python, and I found that the memory usage was an order of magnitude higher than just having a set or list of strings (2mb vs ~100kb), while the performance difference was negligible. I ended up going with a set. The issue is that each layer per node has 26+ instances of your core object struct (whether it's a list or a dict), so any overhead gets ballooned really fast.