Zero-Shot Predictive Models for Relational Databases

ℹ️ EPISODE INFO

Released

August 6, 2026

Duration

5m

Channel

YC Root Access

Watch on YouTube

▶ Open ↗

📝 EPISODE DESCRIPTION

At our inaugural YCML at Startup School, YC Partner Ankit Gupta speaks with Mark Žnidar about predictive models that work directly with relational databases without flattening their tables or manually engineering features. The method represents a relational database as a graph, samples the relevant neighborhood around an entity, and uses specialized attention mechanisms to capture columns, features, and relationships across tables. This preserves information that is often lost in traditional tabular machine learning pipelines. Despite having only 22 million parameters, the model outperformed a four-billion-parameter language model on the item-churn example shown in the presentation. Apply to Y Combinator: https://www.ycombinator.com/apply Work at a startup: https://www.ycombinator.com/jobs

🎤 SPEAKERS

• Ankit Gupta

  host

  Host/interviewer on YC Root Access (Y Combinator).

• Mark Žnidar

  guest

  Researcher/engineer presenting zero-shot predictive modeling for relational databases using transformer-style architectures and specialized attention.

🧾 EPISODE SUMMARY

In this episode of YC Root Access, featuring Ankit Gupta and Mark Žnidar, Zero-Shot Predictive Models for Relational Databases explores zero-shot transformer models predict directly from relational database structures accurately The speaker presents a model that uses relational databases (including foreign/primary key structure) directly rather than flattening tables for traditional feature engineering.

🔗 RELATED EPISODES

Evaluating the Fine-Grained Planning Abilities of Web Agents

ChartNet: Training Vision-Language Models to Understand Charts

Any-Horizon Reasoning for Video Agents

LeanAgent: Lifelong Learning for Formal Theorem Proving

Improving Small Language Model Reasoning With A* Search

Interpretability and Safety for Robot Foundation Models