← All speakers

Bio, Work & Ideas

Andreas Kolleger

Conference affiliation: [REDACTED:username]

Andreas Kolleger, also known professionally as Andreas Kollegger or ABK, leads generative-AI innovation at Neo4j. He helps developers combine knowledge graphs and GraphRAG to give AI applications access to relationships among people, companies, products, and documents that conventional text retrieval can miss.

Earlier in his career, he designed systems for NASA scientific missions and developed medical-informatics systems in Zambia. He was teaching graph databases by 2012, contributed to a Spring Data Neo4j guide, and later served as product manager for Neo4j Bloom, framing graph visualization as a practical tool for exploration, analysis, and communication. His subsequent Neo4j roles included senior developer advocacy before his focus expanded to generative AI.

  • GraphRAG retrieves relationships, not only similar text. His knowledge-graph course teaches developers to combine Cypher, structured relationships, and language models when relevant information is scattered across documents and entities.
  • Agentic knowledge graph construction automates connected context. His more advanced course uses coordinated agents to propose schemas, extract relationships, and connect supplier records, product data, and customer reviews.
  • Developers should shape enterprise AI safeguards. His writing on enterprise AI argues that engineers must influence system architecture, transparency, experimentation, and organizational trust.

At AI Engineer World’s Fair 2024, Kolleger helped facilitate a collaborative GraphRAG workshop exploring vector search, customer-purchase relationships, and personalized recommendations. More recently, his work has extended to context graphs for AI agents.

Read the topics behind these talks

1 conference talk

Key ideas

Scroll to read ↓

Build a retail graph, search product descriptions, personalize results through purchase paths, and use graph embeddings to supply recommendations for a seasonal email.

  • Prepare the notebook and graph database
    0:26 ↗
  • Find an item, personalize it, then suggest what goes with it
    5:59 ↗
  • Build the customer–article–product graph
    11:56 ↗
  • Generate embeddings before creating the index
    26:57 ↗
  • Wrap retrieval without coupling it to one storage layout
    35:47 ↗
  • Derive a peer group from shared purchases
    45:18 ↗
  • Rerank semantic matches without losing cold-start results
    57:39 ↗
  • Turn co-purchase structure into reusable recommendation links
    1:09:33 ↗
  • Refresh embeddings and choose their size
    1:20:07 ↗
  • Use graph queries for questions similarity cannot answer
    1:23:44 ↗
  • Feed two retrieved product lists into one email prompt
    1:29:30 ↗
  • Change the customer, then change the season
    1:37:04 ↗

References