Stop Cramming Documents (Or Chunks) Into One Vector

Hosted by Ben Clavié and Hamel Husain

Tue, Jun 23, 2026

10:30 PM UTC (1 hour)

Virtual (Zoom)

Free to join

Invite your network

Go deeper with a course

Save 25% til Monday
AI Evals For Engineers & PMs
Hamel Husain and Shreya Shankar
View syllabus

What you'll learn

Why single-vector embeddings dilute your data

How pooling a whole document into one vector smooths away the details you actually search for, and the failure modes that show up out-of-domain and when training data is scarce.

How late interaction and MaxSim actually work

A plain-English walk through ColBERT's token-level matching, and why it generalizes so much better than a single averaged vector.

Where retrieval still breaks, and what's next

Ben's honest take on what late interaction still can't do, the gap to perfect retrieval, and how agents are changing what we ask retrieval to do.

Why this topic matters

Almost everyone reaches for single-vector embeddings, then hits the wall: they dilute long documents, fall apart out-of-domain, and need a ton of training data. Ben Clavié built the tools that made late interaction usable (RAGatouille, ModernBERT). He'll show why ColBERT-style retrieval fixes this, how MaxSim works, and where retrieval still falls short, so you know when to reach for it.

You'll learn from

Ben Clavié

Machine Learning Researcher at Mixedbread

Ben Clavié is a machine learning researcher at Mixedbread, focused on NLP and information retrieval. Previously at Answer.AI, he has led projects that broaden access to state-of-the-art retrieval, including RAGatouille and the rerankers library, and co-led the ModernBERT project that modernized the widely used BERT backbone. Based in Tokyo and academically affiliated with the National Institute of Informatics, he also organizes the Late Interaction Workshop series and is exploring what LLMs open up for retrieval.

Hamel Husain

ML Engineer with 25+ years of experience

Hamel Husain is a ML Engineer with over 20 years of experience. He has worked with innovative companies such as Airbnb and GitHub, which included early LLM research used by OpenAI, for code understanding. He has also led and contributed to numerous popular open-source machine-learning tools. Hamel is currently an independent consultant helping companies build AI products.

See all products from Hamel Husain & Shreya Shankar

Sign up to join this lesson

By continuing, you agree to Maven's Terms and Privacy Policy.