Subvolts take Learn how AI models can learn from enterprise databases like graphs, unlocking new prediction capabilities. May 21, 2026 - Developers
- Learn about training neural networks directly on raw multi-table enterprise data and integrating them with agentic systems.
- Business
- Understand how relational foundation models can unlock predictions from your existing databases without extensive retraining.
This interview explores novel AI approaches for scientific discovery and enterprise data, showcasing advanced model capabilities.
Relational Foundation Models for Enterprise Data with Jure Leskovec - #768
Audio streams from Sam Charrington. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
This TWIML AI Podcast episode features Jure Leskovec discussing AI for science and relational deep learning. He covers AI Virtual Cell, which uses single-cell RNA-seq and protein models to understand biology without manual encoding. Leskovec also details Kumo's Relational Foundation Model (RFM2), which treats enterprise databases as graphs for direct neural network training on multi-table data, enabling in-context learning for predictions without retraining.
This episode explores how computers can learn from complex data tables, like those in businesses. It shows how AI can understand relationships within this data to make predictions, similar to how it learns from text, but for structured information.
This episode offers insights into advanced AI applications beyond typical large language models. Listeners will learn how AI can tackle complex scientific problems like protein and cell analysis and how relational foundation models can unlock value in enterprise data. It's for researchers and data scientists interested in novel AI architectures and practical enterprise AI. This work connects to frontier labs' efforts in building more capable and generalizable AI systems that can understand and operate on structured and unstructured data.
- HostSam Charrington
- GuestJure Leskovecco-founder and chief scientist, Kumo
- GuestJure Leskovecprofessor of computer science, Stanford
- Guest: Jure Leskovec
- Title: Relational Foundation Models for Enterprise Data with Jure Leskovec - #768
- Show: The TWIML AI Podcast
- Length: 66 minutes
- AI Virtual Cell uses: single-cell RNA-seq data, protein language models like ESM, and structure models like AlphaFold.
- RFM2 performs: in-context learning over subgraphs to make accurate predictions on new databases and tasks with no training.
In this episode, Jure Leskovec, co-founder and chief scientist at Kumo and professor of computer science at Stanford, joins us to explore two fronts of his work: AI for science and relational deep learning. We begin with AI Virtual Cell, a multiscale effort to learn data-driven representations from proteins to cells to patients using single-cell RNA-seq data, protein language models like ESM, and structure models like AlphaFold—without hand-encoding biology. Jure then dives into relational deep learning, reframing enterprise databases as graphs and training neural networks directly on raw multi-table data. He explains Kumo’s Relational Foundation Model (RFM2), which performs in-context learning over subgraphs to make accurate predictions on new databases and tasks with no training, and how this approach benchmarks against RelBench and other multi-table datasets. We also discuss real-world deployments at companies like Reddit, DoorDash, and Coinbase, explainability via attention over tables and columns, integration with agentic systems, deployment options, and practical limitations.
The complete show notes for this episode can be found at https://twimlai.com/go/768.
- single-cell RNA-seq data
- Jure Leskovec
- Data that shows the activity of genes in individual cells.
- protein language models
- Jure Leskovec
- AI models that understand the sequences and structures of proteins, like language models understand text.
- relational deep learning
- Jure Leskovec
- Using deep learning techniques on data organized in tables with relationships between them.
- in-context learning
- Jure Leskovec
- A model's ability to learn a new task from a few examples given directly in its input, without changing its internal parameters.
53 min1 h 30Why Jev Is Changing How We Build With AI with Diogo Almeida - #779
WithDiogo Almeida
1 h 081 h 05From Voice Agents to AI Avatars with Alexander Smola - #777
WithAlexander Smola
Questions people ask
- What is AI Virtual Cell?
- AI Virtual Cell is a project that uses single-cell RNA-seq data, protein language models, and structure models to learn data-driven representations from proteins to cells to patients without hand-encoding biological knowledge.
- What is Kumo's Relational Foundation Model (RFM2)?
- RFM2 reframes enterprise databases as graphs and trains neural networks directly on raw multi-table data, performing in-context learning over subgraphs for predictions on new databases and tasks without retraining.
Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Sam Charrington.