By Dave DeFusco
Moving digital money is often described as a simple process, like choosing the quickest route on a map, but in today鈥檚 online financial systems, it is far more complicated. In an invited paper to be presented in the main program in February at IEEE ICNC 2026, Katz School M.S. in Artificial Intelligence student Tirth Joshi explains a new way to understand how digital money moves, showing that many financial products are built from other pieces and cannot be reached in a single step. His research, 鈥淗ierarchical Graph Representation for Multi-Chain Blockchain Routing,鈥 introduces a clear, layered model that better reflects how modern digital finance actually works.
In decentralized finance, or DeFi, users often want to move value from one place to another: for example, turning one type of digital asset into another or moving assets across different blockchain networks. Existing systems try to solve this by finding the 鈥渃heapest path,鈥 similar to how a GPS app finds the fastest driving route. Joshi said that this approach can produce invalid routes when dependencies exist in real DeFi systems.
鈥淭he current DeFi routing models are mostly flat,鈥 he said. 鈥淭hey treat everything as if it鈥檚 the same kind of object, and that鈥檚 not how these systems actually work.鈥
One major reason is that many digital assets are built from other assets. Some assets are simple, like basic digital coins. Others are more complex and only exist if several simpler assets are combined in the right way. For example, a popular type of asset called a liquidity pool share can only be created if you already have two different coins, often in a specific ratio.
鈥淚f you ignore how these assets are built,鈥 said Joshi, 鈥渢he system might suggest a route that looks cheap, but it鈥檚 actually impossible to follow.鈥
To make this easier to understand, Joshi uses a simple metaphor. He compares blockchain systems to buildings. Each building represents a separate blockchain network. Inside each building are floors. The ground floor holds basic assets. Higher floors contain more complex assets that depend on what鈥檚 below them.
鈥淵ou can鈥檛 reach the higher floors without passing through the lower ones,鈥 said Joshi. 鈥淭hat鈥檚 how many financial products work in DeFi.鈥
This building-and-floor model helps show the difference between two kinds of decisions. Some steps are choices: pick one option or another. Other steps are requirements: several things must all be done before you can move forward. Joshi calls these 鈥淥R鈥 and 鈥淎ND鈥 decisions.
A swap can be thought of as an OR choice, but creating a complex asset like vault tokens, LP shares and wrapped assets is an AND decision because one must have all required pieces first. Traditional systems do not handle this difference well. In computer science, problems that involve many AND decisions are known to be extremely hard to solve perfectly; however, Joshi made an important observation: in real DeFi systems, these complicated dependency chains are usually short.
鈥淲hen we looked at real examples,鈥 he said, 鈥渨e found that you rarely have many layers of these requirements stacked on top of each other.鈥
Using this insight, Joshi designed a new method called the GHP algorithm. Instead of trying to solve everything at once, the algorithm works step by step, reusing earlier results and avoiding unnecessary repetition. This makes it fast while still producing answers that are very close to the best possible solution. In tests on tens of thousands of tokens across dozens of chains showed that the new method ran more than eight times faster than traditional approaches. At the same time, its results were within 5% of the most exact, but much slower solutions.
Honggang Wang, chair of the Department of Graduate Computer Science and Engineering, said the research addresses a growing real-world problem.
鈥淎s blockchain systems become more complex and interconnected, we need models that reflect how these systems truly operate,鈥 he said. 鈥淭his work provides a clear and scalable way to think about routing in modern decentralized finance.鈥
Joshi emphasized that the project would not have been possible without Wang鈥檚 mentorship. 鈥淚鈥檓 very grateful to Professor Wang for his support and guidance throughout this work,鈥 he said.
Looking ahead, Joshi plans to continue this research by accounting for real-time changes such as transaction fees, delays and reliability when moving assets between blockchains.
鈥淚n this space, speed matters,鈥 said Joshi. 鈥淧rices and conditions change constantly. A solution that鈥檚 fast and good enough is often more useful than one that鈥檚 perfect but too slow.鈥
