Oshani Seneviratne and Fernando Spadea
Rensselaer Polytechnic Institute
Title: Predicting Risk in Web3: From Crypto Social Signals to DeFi Lending Outcomes
Abstract: Cryptocurrencies and decentralized finance generate massive, real-time data streams–ranging from social media discourse to immutable blockchain transaction logs. A central challenge (and opportunity) is prediction: can we anticipate major market events, shifts in user behavior, and risk outcomes early enough to inform action?
In this talk, we present three connected case studies that move from modeling goals to AI methods to financial applications. First, in Deciphering Crypto Twitter, we ask whether social signals can forecast or contextualize real-world crypto incidents. Using a dataset of 40 million tweets, we apply sentence-level embeddings, clustering, and network analysis to uncover thematic communities, coordinated behavior, and sentiment shifts that align with major events such as the November 2022 FTX collapse.
Second, in Benchmarking Temporal Web3 Intelligence, we focus on predicting how users evolve over time in decentralized systems. We introduce the lessons learned from the FinSurvival 2025 Challenge, built on 21.8 million Aave v3 transactions and 16 time-to-event prediction tasks. We summarize what worked in the winning approaches and highlight practical lessons for building reproducible temporal benchmarks in fast-changing environments.
Finally, in Actionable Risk Intelligence for DeFi Lending Protocols, we move from prediction to intervention. We describe an autonomous agent that estimates liquidation risk, simulates counterfactual futures, and identifies minimal-capital actions to reduce liquidation likelihood. Using a protocol-faithful Aave v3 simulator on 5,012 user profiles, the agent substantially reduces liquidations while distinguishing true insolvency risk from routine “dust” events.
Overall, the talk demonstrates how AI methods, applied to both social and transactional data, can deliver practical risk intelligence for modern financial platforms, while raising new research questions at the intersection of machine learning, dynamic networks, and decentralized systems.
Oshani Seneviratne Bio: Oshani Seneviratne is an Assistant Professor of Computer Science at Rensselaer Polytechnic Institute (RPI), where she leads the BRAINS Lab (Bridging Resilient, Accountable, Intelligent Networked Systems). Her research focuses on decentralized systems, including web technologies, blockchain, and decentralized learning, with applications in health informatics and fintech. Her research has been recognized with multiple best paper awards and the Yahoo! Key Scientific Challenges Award. For more information about Oshani, please refer to her website http://oshani.info.
Fernando Spadea Bio: Fernando is a fourth-year CS PhD student at RPI advised by Prof. Oshani Seneviratne. He is doing research in recommendation systems in multiple domains, including finance, with an emphasis on decentralized systems. His prior and current research has had him working with open-weight large language models, financial data, blockchain, Solid, federated learning, and privacy guarantees.
Date: Wednesday, March 11, 2026, 3:30 PM, AUST 434
WebEx link: https://uconn-cmr.webex.com/uconn-cmr/j.php?MTID=m22ab507cb9f46e2ce2fab499c49a5f9a
Coffee will be available at 3:00 PM in the Noether Lounge (AUST 326)
For more information, contact: Yuwen Gu at yuwen.gu@uconn.edu