A Pragmatist’s Guide to Building Knowledge Graphs from Unstructured Data

🎥 Recorded live at the MLOps World | GenAI Summit 2025 — Austin, TX (October 8, 2025) Session Title: I Tried Everything: A Pragmatist’s Guide to Building Knowledge Graphs from Unstructured Data Speaker: Alessandro Pireno, Founder, Stealth Company Talk Track: Data Engineering in an LLM Era Abstract: Traditional ETL pipelines are collapsing under the demands of LLMs and unstructured data. While they excel at handling structured tables, they fail to capture the context and implicit relationships that modern AI systems depend on. In this session, Alessandro Pireno introduces a pragmatic new approach: KG-ETL — pipelines that treat knowledge graphs as a first-class output. He breaks down three competing architectures for constructing knowledge graphs from raw data and reveals how to use LLM prompts as a new “T” in your pipeline. Through hands-on comparison, Alessandro contrasts LLM-based, traditional NLP-based, and hybrid vector search-based architectures — showing how hybrid retrieval workflows use vector stores for entity resolution far beyond semantic search. What you’ll learn: • Design and compare three data pipeline architectures for knowledge graph construction • Evaluate cost, latency, scalability, and observability trade-offs for each approach • Learn a novel vector-store technique for high-precision entity resolution using FastText embeddings • Apply a decision framework to select the right KG-ETL pipeline for your data type and business needs.