Intro
Subscribe
Bridging Knowledge Graphs and AI: A New Era
Recent advancements in Large Language Models (LLMs) are revolutionizing knowledge graph capabilities. The emergence of cost-effective, fine-tunable open-source LLMs in the cloud, alongside powerful closed-source deep research tools from OpenAI, Google, and Anthropic, democratizes access. Even small teams can now afford to create exceptionally powerful and sophisticated knowledge systems, transforming data into actionable insights with unprecedented ease and efficiency.

Below are a couple of projects that are production ready and open source. We are using both in our tech stack.
PRefLexOR: Enhancing Knowledge Graph Representations
Developed by MIT's Laboratory for Atomistic & Molecular Mechanics, PRefLexOR (Pre-trained RDF Lexical Representation) is an innovative project that generates powerful lexical embeddings for RDF datasets. By capturing the semantic meaning of words and phrases within knowledge graphs, PRefLexOR creates richer representations of entities and relationships. This cutting-edge approach significantly boosts the accuracy of tasks like link prediction and entity alignment, making modern knowledge graphs more robust and precise for advanced AI applications.



Knowledge Crystals
The PRefLexOR paper introduces the concept of Knowledge Crystals.

A knowledge crystal is a highly organized, validated, and reusable package of knowledge that combines data, context, and reasoning into a single, clear, and actionable resource. Unlike ordinary knowledge graphs that simply link data points, knowledge crystals capture not only facts but also the logic, evidence, and trustworthiness behind those facts.
We are actively working to apply this research to other fields of science and other problem domains such as business intelligence.
Why Knowledge Crystals Matter for Business
Knowledge Crystals transform raw data into a strategic asset for enterprises. They empower organizations to leverage their information with unprecedented clarity, reliability, and agility, driving superior decision-making across all functions.
Validated & Reliable
Each crystal is meticulously checked to ensure accuracy, significantly reducing misinformation and guesswork in decision-making processes.
Contextual & Explainable
They don’t just show data relationships—they explain the "why" and "how," providing clear reasoning paths for complex insights.
Modular & Scalable
Built as independent, reusable units, knowledge crystals allow for easy combination and growth of your knowledge base without losing clarity.
Actionable Insights
By linking data with robust reasoning and evidence, teams can make faster, smarter decisions supported by transparent logic.
Cross-Domain Integration
Seamlessly integrates knowledge from different departments, breaking down silos and enabling truly holistic organizational insights.
Zep/Graphiti: Persistent Memory for LLMs
Zep, powered by Graphiti, offers a robust memory system for Large Language Models, overcoming their stateless nature. This innovative solution enables LLMs to retain and recall conversation history, user preferences, and real-world context across sessions. By transforming conversational data into a knowledge graph, Zep allows LLMs to leverage rich, structured information for more coherent, personalized, and insightful interactions.

We can use Graphiti to store many PRefLexOR or other types of graphs such as the json-ld format generated by our Deep Research Protocol. This makes the entire corpus efficient for humans or LLM's to search, edit and many other tasks. Connecting to a Graphiti data store grounds LLM's to reduce mistakes. The Graphiti data store essentially makes any LLM and expert on the knowledge stored in the individual graphs and the corpus as a whole.
Deep Research: Powering Knowledge Extraction
Learn how to use Deep Research to extract knowledge graphs and how to use knowledge graphs you create to power Deep Research.