Jo Kristian Bergum

Co-founder of Hornet, a retrieval infrastructure company focused on agentic AI systems. Formerly Chief Scientist at Vespa AI, where he led research on AI-powered search and retrieval, including the native ColBERT embedder and LLM-as-judge work.

Articles in This Vault

• Announcing the Vespa ColBERT Embedder
• Using Approximate Nearest Neighbor Search to Find Similar Products
• E-commerce Search and Recommendation with Vespa
• Improving retrieval with LLM-as-a-judge
• This Is What Agentic Retrieval Looks Like — Hornet; GPT-5 query behavior analysis on BrowseComp-Plus

Key Contributions

• Native ColBERT embedder in Vespa with 32x asymmetric compression
• Asymmetric binarization (float queries, int8 documents) for late interaction at scale
• LLM-as-judge retrieval evaluation framework (Vespa)
• Characterization of agentic query workload: long queries, web-search operators, multi-turn compounding (Hornet)

Concepts

• ColBERT
• Late Interaction
• Dense Vector Retrieval
• Hybrid Search
• Agentic Retrieval
• LLM as Judge