I’m excited (and a bit nervous) to share the second episode of The Executive Code podcast, where my colleague Kirim and I explore the intersection of AI, engineering, and leadership. This new podcast project is something I’ve been working on with the team at Snapshot AI and Flatiron Software .
I feel a little intimidated jumping into such deeply technical AI discussions on camera. When I think of my friends who are real-deal AI experts like Richard Socher , Bryan McCann , Jeremy Howard , Vipul Ved Prakash , and my brother Vik , a longtime AI expert who has a PhD. in a related field, I can’t help but feel like something of an imposter discussing these deeply technical AI concepts. But that’s why we started this podcast. Learning and teaching go together.
In this episode, Kirim and I discuss the breakthrough RAPTOR paper from Christopher Manning and his colleagues at Stanford University . We break down how this innovative approach changes the way AI models retrieve and understand information from long documents—with significant implications for how companies like ours approach developer data analysis.
What’s RAPTOR and Why Does It Matter?
RAPTOR (Recursive Abstractive Processing for Tree-Organized Retrieval) addresses one of the most significant challenges in modern AI: helping language models effectively understand and retrieve information from lengthy documents. The traditional approach of simply breaking documents into chunks misses crucial context and relationships between ideas.
Instead, RAPTOR builds what Manning calls a “hierarchical tree of summaries”—organizing information at different levels of detail. This mirrors how humans naturally think, allowing AI systems to understand both specific details and broader context simultaneously.
In the podcast, Kirim and I discuss:
- How RAPTOR processes documents by clustering similar chunks and creating summaries at multiple levels
- The impressive performance improvements (up to 20 percentage points) when combined with models like GPT-4
- Why this approach scales better with document length than traditional methods
- Practical applications for development teams and engineering organizations
Applications for Engineering Teams
For those building or leading engineering teams, RAPTOR’s approach has exciting implications:
- Repository organization: Creating hierarchical representations of codebases from individual functions up to system architectures
- Development timelines: Clustering related work items and building summaries at different time scales
- Knowledge transfer: Enabling new team members to ask questions and get context at the right level of detail
The synergy between these techniques and larger context models like Llama 4 could transform how teams understand their development processes—moving beyond surface-level metrics to genuine comprehension of complex systems.
Thanks To
I couldn’t have done my part in this project without several people. First, my co-host Kirim, whose technical expertise and entrepreneurial mindset make our conversations so valuable. I’m also deeply grateful to our colleague Ana Clara for her expert guidance and brutally candid feedback that shaped every aspect of this project.
And of course, a special thanks to Christopher Manning and his co-authors for their innovative research that’s making a real impact on how we build AI systems.
Join the Conversation
If you found this discussion valuable, I’d be grateful if you would share your thoughts in the comments on the YouTube video . Your comments help the YouTube algorithm understand who might benefit from this content, and more importantly, they help us improve future episodes.
We’ll be releasing new episodes every other week, exploring the frontiers of technology and leadership. Subscribe to stay updated, and feel free to suggest topics you’d like us to cover in future episodes.
The Executive Code is filmed across New York, London, and wherever the conversation takes us.