Hydro, Datathink and General Purpose Programming

Bio: Joseph M. Hellerstein's work focuses on data-centric systems and the way they drive computing. He is the Jim Gray Professor of the Graduate School at the University of California, Berkeley, and a Distinguished Scientist at Amazon Web Services. Recognitions of his research contributions include the ACM SIGMOD Codd Innovations Award, the ACM Fellow and Alfred P. Sloan Research Fellow awards, and seven "Test of Time" awards for his research papers. He co-founded RunLLM, a vendor of AI assistants, and Trifacta, the pioneering company in Data Wrangling. Hellerstein has taught thousands of Berkeley students and advised over 30 Ph.D.'s, three of whom were recognized with the ACM SIGMOD Jim Gray Dissertation award.

SagaLLM: Context Management, Validation, and Transaction Guarantees for Multi-Agent LLM Planning

Abstract: This talk introduces SagaLLM, a structured multi-agent architecture designed to address four foundational limitations of current LLM-based planning systems: unreliable self-validation, context loss, lack of transactional safeguards, and insufficient inter-agent coordination. While recent frameworks leverage LLMs for task decomposition and multi-agent communication, they often fail to ensure consistency, rollback, or constraint satisfaction across distributed workflows. SagaLLM bridges this gap by integrating the Saga transactional pattern with persistent memory, automated compensation, and independent validation agents. It leverages LLMs' generative reasoning to automate key tasks traditionally requiring hand-coded coordination logic, including state tracking, dependency analysis, log schema generation, and recovery orchestration. Although SagaLLM relaxes strict ACID guarantees, it ensures workflow-wide consistency and recovery through modular checkpointing and compensable execution. Empirical evaluations across planning domains demonstrate that standalone LLMs frequently violate interdependent constraints or fail to recover from disruptions. In contrast, SagaLLM achieves significant improvements in consistency, validation accuracy, and adaptive coordination under uncertainty—establishing a robust foundation for real-world, scalable LLM-based multi-agent systems.

Bio: Edward Y. Chang is an adjunct professor of Computer Science at Stanford University (2019–present) and co–Editor-in-Chief of ACM Books (effective Dec. 2025). He previously served as President of HTC DeepQ Healthcare (2012–2021) and as a Director of Research at Google (2006–2012), where he led work on scalable machine learning, web-scale image annotation, and data-centric AI, and sponsored the ImageNet project. Earlier, he was a tenured faculty member in ECE at UC Santa Barbara (1999–2006) and a visiting professor at UC Berkeley (2017–2020). Chang earned an M.S. in Computer Science and a Ph.D. in Electrical Engineering from Stanford. He is an ACM and IEEE Fellow, with honors including the NSF CAREER Award, Google Innovation Award, ACM SIGMM Test of Time Award, and a US$1M XPRIZE for AI-driven diagnosis. He is the author of several books, most recently Multi-LLM Agent Collaborative Intelligence: The Path to Artificial General Intelligence (March 2024; revised October 2025).

LLM-powered Data Tooling: the Next Frontier

Abstract: LLMs are changing the world, but how can they help with data processing?  In this talk, we discuss ongoing work in the EPIC Data Lab at Berkeley to rethink the end-to-end data lifecycle, now with LLMs in the mix.  We describe our scalable, efficient, and usable text data processing system stack, aka our document "stack stack", as well as a couple of projects that are having impact across a number of real-world domains.  We'll also briefly touch upon our future research vision around better supporting agentic workloads.

Bio: Aditya Parameswaran is an Associate Professor in EECS at UC Berkeley, and a co-director of the EPIC Data Lab. Aditya has published 100+ papers overall at top venues across multiple disciplines, with multiple best paper awards;  just this year, his papers have appeared at the top DB (VLDB, SIGMOD), AI (ICLR, NAACL), and HCI (UIST, CSCW) venues.  Multiple open-source tools developed in his group have received thousands of GitHub stars (including Modin, Lux, IPyFlow, DocETL)---and have been downloaded tens of millions of times overall across a spectrum of industries.  His research was commercialized as a startup, Ponder, in 2021, where he served as Co-founder and President, before its acquisition by Snowflake. Aditya has received the Alfred P. Sloan Research Fellowship, VLDB Early Career Award, the NSF CAREER Award, the TCDE Rising Star Award, along with other recognitions. His website is at http://adityagp.net

Scalable Secure-Oblivious Computation for Next Generation Private Systems, Databases, and AI 

Abstract: Any privacy-preserving computation on encrypted data that relies solely on encryption can leak significant information about the plaintext input through leakage-abuse attacks. Industrial approaches that support confidential computing through hardware enclaves are susceptible to side-channel attacks; however, hardware enclaves provide an affordable and low-cost solution for any privacy-preserving computation. Oblivious primitives are a powerful cryptographic tool that, when combined with hardware enclaves, can mitigate leakage-abuse and software side-channel attacks.   

Oblivious primitives find applications in various areas, including Signal’s contact discovery, Anonymous Key Transparency, end-to-end encrypted email search, differential privacy in the shuffle model, large-scale software monitoring (e.g., Google’s Prochlo), private federated learning/computation (Apple’s Private Cloud Compute), Data Clean Rooms, LLM privacy/private RAG, Google’s Privacy Sandbox, and broader confidential computing efforts (including Oracle’s). In this short-talk, we discuss our recent hardware-enclave-based oblivious primitives that scale private computations to terabyte-sized inputs—far exceeding the previous state-of-the-art 100MB-4GB range. Our scalable oblivious primitives include a high-throughput oblivious key-value store (used in Signal’s contact discovery, SOSP’21), a low-latency approach (PVLDB’24), the most scalable oblivious sort and shuffle (IEEE SP’24), and the first scalable oblivious filter, group-by, join approaches (USENIX’25).  

Bio: Ioannis Demertzis is an Assistant Professor in the Computer Science and Engineering Dept. at the University of California, Santa Cruz. His research focuses on applied cryptography, security & privacy, and secure databases/systems. His work has been published at top security, system and database conferences including USENIX, CRYPTO, NDSS,  S&P, SIGMOD, SOSP, PVLDB and TODS. He is the recipient of the ACM SIGSAC Doctoral Dissertation Award Runner-up, Distinguished Dissertation Award of ECE (University of Maryland), and the Symantec Research Labs Graduate Fellowship. Before joining UCSC, he was a Postdoctoral Researcher at the EECS Dept. of UC Berkeley hosted by Prof. Raluca Ada Popa. He received his Ph.D. from the ECE Dept. of the University of Maryland, College Park advised by Prof. Charalampos Papamanthou. He obtained his ECE Diploma and M.Sc at the Technical University of Crete, under the supervision of Minos Garofalakis.

C. Mohan (Moderator) 

Dr. C. Mohan is Distinguished Professor of Science at Hong Kong Baptist University, Distinguished Visiting Professor at Tsinghua University, and a member of the inaugural Board of Governors of Digital University Kerala. He retired in 2020 as an IBM Fellow after 38.5 years at IBM Almaden Research Center, where his pioneering work in databases, blockchain, AI, and related areas influenced numerous products, standards, and the academic community. He is best known for inventing the ARIES database recovery algorithms and the Presumed Abort commit protocol.An IBM, ACM, and IEEE Fellow, Mohan has received honors including the ACM SIGMOD Edgar F. Codd Innovations Award, VLDB 10-Year Best Paper Award, and election to both the US and Indian National Academies of Engineering. He holds 50 patents and has served as consultant to Google and Microsoft. A prolific speaker, he has delivered talks in 43 countries and served on numerous editorial and advisory boards. More details at https://bit.ly/CMwIkP and his homepage at https://bit.ly/CMoDUK.

Dhruba Borthakur (Open AI) 

Dhruba Borthakur is the Technical Lead of the data infrastructure team at OpenAI. He co-founded Rockset, a search database that powers AI applications at OpenAI.  Dhruba was the founding engineer of the RocksDB database at Facebook and one of the founding engineers of the Hadoop File System at Yahoo. Dhruba was also an early contributor to the open source Apache HBase project. Previously, he was a senior engineer at Veritas Software, where he was part of a team responsible for the development of VxFS and Veritas SanPointDirect storage system; was the cofounder of Oreceipt.com, an ecommerce startup based in Sunnyvale; and was a senior engineer at IBM-Transarc Labs, where he contributed to the development of Andrew File System (AFS). Dhruba holds an MS in computer science from the University of Wisconsin-Madison and a BS in computer science BITS, Pilani, India. He has 69 issued patents.

Matei Zaharia (Databricks & UC Berkeley)

Matei Zaharia is an Associate Professor of EECS at Berkeley and a Cofounder and CTO of Databricks. He started the Apache Spark open source project during his PhD at UC Berkeley in 2009, and has worked broadly on other widely used data and AI software, including Delta Lake, MLflow, Dolly and ColBERT. He currently works on a variety of research projects in cloud computing, database management, AI and information retrieval. Matei’s research was recognized through the 2014 ACM Doctoral Dissertation Award, an NSF CAREER Award, and the US Presidential Early Career Award for Scientists and Engineers (PECASE).

Yannis Papakonstantinou (Google)

Yannis is a Google Cloud Distinguished Engineer in the Data Cloud organization. He technically leads innovative features and products in support of GenAI and agentic applications over Data Cloud services. In addition he advises many initiatives at the intersection of GenAI and databases. He is also an Adjunct Professor of Computer Science at UCSD and has held roles at Databricks, AWS, and as CEO/Chief Scientist of Enosys Software, which developed an early Enterprise Information Integration platform later acquired by BEA. A former Stanford Ph.D. in Computer Science, his research spans query processing, semistructured data, and AI, with over 120 publications and 21,000+ citations.

Avrilia Floratou (Microsoft)

Avrilia Floratou is a Principal Engineering Manager in Microsoft Azure Data, where she leads the development of AI components for the Fabric data analytics platform. Her current work centers on building intelligent agents that enable business users to derive insights from their data, as well as designing services for AI-driven data transformation and extraction.

Previously, she led a research team at Microsoft’s Gray Systems Lab, exploring topics ranging from applying AI to data management to improving system performance and scalability. In this role, she collaborated closely with product teams like Fabric and SQL Server to translate research advances into real-world product impact. Prior to Microsoft, she spent three years at IBM Almaden Research Center, contributing to research on SQL-on-Hadoop and scalable analytics.

Avrilia holds a PhD and MS in Computer Science from the University of Wisconsin–Madison and a Bachelor’s degree from the University of Athens.

Shasank Chavan (Oracle)

Shasank Chavan is the Vice President of the In-Memory, Data and AI Engines group at Oracle. He leads an organization of brilliant engineers working on the nexus between AI systems and modern databases. His team is responsible for developing the next-generation AI-centric data storage engine, designed for in-memory OLTP, Analytics and AI Vector Search capabilities to power real-time AI intelligence in database systems.  Shasank earned his BS/MS in Computer Science at the University of California, San Diego. He has accumulated 50+ patents over a span of 27 years working on systems software technology.