Center for Digital Finance and Technology Announces New Research Grants
The Center for Digital Finance and Technology (CDFT) at Columbia Engineering is pleased to announce a fourth year of CDFT Research Grants. Established in 2022, the Center is focused on advancing the digital transformation of financial services for higher efficiency and security, increased accessibility, and greater social responsibility.
The projects selected this year tackle state-of the art methodologies for automated market makers as well as new proposed technologies in decentralized systems for privacy, security, and performance.
Awarded proposals are:
TrailFI: Safe and Auditable Agentic Automation for Fintech
PI: Zhuo Zhang, Assistant Professor, Department of Computer Science, Columbia University
Autonomous financial agents are moving from giving advice to moving money. They can monitor markets, call tools, control wallets, and submit transactions that may be hard to reverse. But when something goes wrong, today's records often leave a dangerous gap. Chat logs may show what the agent said. Blockchain records may show what happened. Neither clearly explains why the agent took the action. This project propses TrailFI to closes this gap. TrailFI captures the agent’s key decision context, links it with tool, wallet, system, and blockchain evidence, and reconstructs the causal path after an incident. The goal is to turn scattered technical traces into a clear, evidence-backed explanation. With TrailFI, risk teams can understand what went wrong, engineers can verify the cause, and autonomous financial systems can become safer, more accountable, and more ready for real- world deployment.
Evaluating and Improving the Stability of Financial Networks through Cooperative Game Theory and Optimization
PI: Yuri Faenza, Associate Professor, Department of Industrial Engineering and Operations Research, Columbia University
The increasing transparency of digital financial systems offers a unique opportunity to redefine systemic risk management. In this proposal, we aim to exploit this research opportunity to propose a novel investigation of the stability of financial networks using tools from Cooperative Game Theory, and for suggesting possible interventions by a Central Authority using tools from Optimization. Our proposal distinguishes itself from the literature in the use of solution concepts not previously investigated in this context, in the operational approach to the role of the Central Authority, and in the incorporation of features that are specific to Digital Finance.
Efficient Branch-and-Bound Verification of Zero-Knowledge Virtual Machines
PI: Junfeng Yang, Professor, Department of Computer Science, Columbia University
Zero-Knowledge Virtual Machines (zkVMs) enable verifiable execution of general-purpose programs by translating semantics into algebraic constraints. Correctness is critical; missing constraints allow forged proofs, directly threatening the economic security and stability of decentralized financial protocols. Existing verification methods—including our prior CDFT-funded system, ZKFUZZ —fail to scale or guarantee completeness. We propose ZEBRA, an automated verification framework reducing zkVM correctness to a solution-set cardinality problem over a canonical trace space. By leveraging the structural sparsity of zkVM constraints (averaging 14.0% connectivity), ZEBRA uses interval-based branch-and-bound search to either produce concrete counterexamples or certify the absence of vulnerabilities. Our prototype has already discovered 11 zero-day flaws. This project will deliver scalable verification tools to secure foundational digital finance infrastructure.
Competitive Fee Design for Decentralized Exchanges
PI: Marcel Nutz, Professor, Department of Statistics, Columbia University
This project develops a theory of competitive fee design for automated market makers (AMMs), extending our CDFT-supported work on AMM fees and liquidity provider (LP) profitability. We will study how multiple decentralized exchanges (DEXs) compete for order flow in the presence of a centralized exchange (CEX), and how this competition shapes volume, arbitrage losses, fee revenue, and long-run LP returns. The project combines a multi-exchange simulation model with a tractable continuous-time approximation that yields explicit formulas for these quantities. We will characterize when equilibrium forces produce fee undercutting or market segmentation, and how CEX trading costs influence DEX fee levels. The deliverables include analytical and numerical results, as well as open-source simulation code.
Trustworthy Financial Agents Based on Blockchain and Zero-Knowledge Proofs
PI: Xiaodong Wang, Professor, Department of Electrical Engineering, Columbia University
Financial large language models (LLMs) are rapidly evolving from chat-based systems into autonomous financial agents capable of analyzing SEC filings, generating investment reports, invoking external tools, and supporting compliance and trading workflows. These capabilities create significant opportunities for digital investment, compliance automation, and risk management, but they also introduce new challenges: unsupported financial claims, unverifiable computations, unauthorized actions, and insufficient auditability. This project develops a prototype framework for trustworthy financial agents powered by Fin-GPT. The key idea is to move from unconstrained agents to contract-governed and verifiable execution. Financial agents generate structured execution traces that record prompts, retrieved evidence, tool calls, computations, and candidate outputs. Smart contracts then enforce evidence, computation, authorization, and audit rules before outputs are released. Blockchain infrastructure anchors immutable commitments to execution traces, while zero-knowledge proofs (ZKPs) enable selected computations and policy checks to be verified without exposing sensitive financial data. The project aligns with CDFT priorities in smart-contract design, security and privacy of agentic financial systems, agentic automation for FinTech, digital investment infrastructure, and smart-contract risk analysis. By the end, we will deliver a trace-generating agent prototype, a smart- contract governance layer, and a ZKP verification demo for privacy-preserving financial reasoning.