Navigating a Full Stack Knowledge Architecture from Database to LLM
summary
Claims a reproducible pipeline that turns SEC Edgar 10‑Ks and 13F filings into a layered knowledge graph, using Python + LangChain to chunk text, create vector embeddings, run vector search and mix that with structured queries over new forj via Cipher. Demo shows generated Cypher queries sometimes work and sometimes hallucinate, evaluation (manual unit tests) uncovers failure modes, and the system surfaces investors and risk signals but needs query vetting, chunking choices, and indexing tweaks to reach reliable results.
AI-written summary · about
chapters (9)
- 0:00Full stack knowledge intro
- 3:04LLM limits and RAG basics
- 7:37How retrieval augmented generation works in practice
- 10:44Knowledge stack: DB to LLM
- 16:44Knowledge graph design
- 23:20Chunking and embeddings
- 29:17Unstructured + structured data
- 31:44Live demo: graph integration
- 40:34Evaluating LLM responses
related talks
44mGive Your LLMs a Left Brain
20mPower Amazon Bedrock applications with Neo4j knowledge graph
01h 59mLLM applications with LangChain
38mKnowledge Graphs : Adding a New Dimension to Vector Search and Generative AI
25mLangChain with Azure AI: Building LLM powered Applications
30mBuilding Knowledge Graphs With LLMs
23mConnecting the Dots With Context Graphs
22mBuilding Cypher Queries With Open-Source LLMs