Clyep

12 episodes, about 1 hr 15 min 2 views across the series

Topics: Retrieval and knowledge

Learn Knowledge Graphs

Created by Abhyuday DesaiUpdated , episode 4

Knowledge graphs from the ground up: what an ontology declares, how a system decides that two records are one thing, and when a graph beats vector search in front of an LLM.

Start with episode 1

About this series

Twelve short videos, four to seven minutes each, on modeling a domain explicitly and using that model for retrieval. The first part builds the vocabulary: entities, attributes, relationships and triples, what an ontology declares that a knowledge graph does not, and when explicit structure earns its cost. The second part solves identity: how a system decides that several records describe one real company, and what a wrong answer does to a graph. The third part spends that identity: what only a graph can answer, how documents and a graph work together, and how a language model sits on top.

Every record, score and query on screen is illustrative. Nothing is executed and there is no companion repository, by design: the videos teach the decisions, not a product.

Part oneModeling: ontologies and knowledge graphs

  1. 01

    Entities, Attributes, and Relationships

    What a knowledge graph is made of: entities with identities, attributes as values, relationships as typed edges, and a fact written down as a triple.

    6:40 0 views Published
    Watch
  2. 02

    What an Ontology Declares

    The ontology is the rulebook: types, which relationships are allowed between them, and how many values a thing may carry. The knowledge graph is what has been recorded under those rules.

    5:39 0 views Published
    Watch
  3. 03

    What a Knowledge Graph Doesn't Prove

    An edge in a knowledge graph is a recorded claim, not a verified one. What assertions establish, where the semantic layer fits, and the three conditions that make explicit structure worth maintaining.

    4:41 0 views Published
    Watch

Part twoIdentity: entity resolution

  1. 04

    Entity Resolution: Why Deduplication Isn't Enough

    Three systems hold three spellings of one company. Why distinct-by-string can't fix that, the two opposite ways identity goes wrong, and the two cheapest steps of a real resolution pipeline.

    5:46 1 view Updated , v2
    Watch
  2. 05

    Entity Resolution: Blocking and Scoring

    Blocking cuts comparisons to plausible pairs; scoring turns each pair's evidence into one number. Why one pass misses matches, why missing and conflicting evidence differ, and where embeddings fit.

    6:48 0 views Published
    Watch
  3. 06

    Entity Resolution: From Score to Identity

    A score becomes a decision at two cut points, and a decision becomes an identity through review, survivorship, provenance and reversible merges.

    8:00 0 views Published
    Watch
  4. 07

    What a Wrong Identity Does to a Knowledge Graph

    Identity failures seen from inside the graph: a split identity breaks a traversal, resolving it reconnects the path without adding a fact, and an incorrect merge produces a confidently wrong answer.

    5:29 0 views Published
    Watch

Part threeRetrieval: graphs, vectors and LLMs

  1. 08

    A Similarity Score Is Not a Relationship

    A similarity score says two texts are close. A graph edge says what connects two entities and in which direction. The difference, and a two-hop question answered by joining recorded relationships.

    8:18 0 views Published
    Watch
  2. 09

    What Only a Knowledge Graph Can Answer

    The questions only structure answers: an absent relationship, a complete set rather than a top-k, a count of connections. And the honest case where a sentence nobody modeled beats the graph.

    6:02 0 views Published
    Watch
  3. 10

    Documents and the Knowledge Graph, Together

    Two links join prose to a graph: extraction and attachment. Then the two ways a vector database and a knowledge graph run together: similarity finds the entry point, structure bounds the search.

    6:51 0 views Published
    Watch
  4. 11

    Grounding an LLM in a Knowledge Graph

    Two ways to put a language model on a knowledge graph: let it write a query from the schema, or hand it the facts with identifiers intact. And three things grounding does not guarantee.

    5:14 0 views Published
    Watch
  5. 12

    Four Questions Before You Build a Knowledge Graph

    What the heaviest form of graph structure costs to keep current, and four questions that settle the choice between a knowledge graph and a vector database. The series close.

    5:23 1 view Published
    Watch