WhyHow.AI

Problem Statement: How do we choose the right document from thousands of our documents to retrieve information from in a RAG system? Open-Source Knowledge Table as a Document Metadata generation tool for File Directories. When we at WhyHow.AI were dealing with enterprise systems with thousands of documents, the first most obvious problem to solve for is about which of the documents are appropriate to answer the question, before we then look at the information in those documents to construct the answer in a 2-Step RAG process. This pattern is fairly common, but the biggest restriction here is in generating appropriate labels for documents. With Knowledge Table, one can easily generate and save custom and common metadata of your documents, and saving it in memory with just a single click. Chris Rec Thomas Smoker

What's better than GraphRAG? StructRAG! Imagine asking your AI assistant a complex question that requires piecing together information from multiple sources. With current methods, the AI often struggles to find and connect the relevant bits of information scattered across various documents. But what if the AI could automatically organize that raw information into a clear knowledge structure optimized for the task at hand? That's the key idea behind StructRAG, a new framework that aims to replicate how humans tackle knowledge-intensive reasoning. And here's how it works: 1. StructRAG first identifies the best way to structure the knowledge for the specific task, such as a table, graph, or tree. 2. It then reconstructs the original documents into this structured format, making it easier to see connections and relationships between pieces of information. 3. Finally, it uses this structured knowledge to infer the answer to the original question. StructRAG outperforms existing methods on a range of challenging reasoning tasks that require combining multiple facts and insights. For example, it excels at open-ended science questions that require understanding complex processes and piecing together evidence from multiple studies. By mimicking how humans organize information to solve problems, they were able to match - or even surpass - GraphRAG while being much faster. ↓ Liked this post? Join my newsletter with 50k+ readers that breaks down all you need to know about the latest LLM research: llmwatch.com ??

What's better than GraphRAG? StructRAG! Imagine asking your AI assistant a complex question that requires piecing together information from multiple sources. With current methods, the AI often struggles to find and connect the relevant bits of information scattered across various documents. But what if the AI could automatically organize that raw information into a clear knowledge structure optimized for the task at hand? That's the key idea behind StructRAG, a new framework that aims to replicate how humans tackle knowledge-intensive reasoning. And here's how it works: 1. StructRAG first identifies the best way to structure the knowledge for the specific task, such as a table, graph, or tree. 2. It then reconstructs the original documents into this structured format, making it easier to see connections and relationships between pieces of information. 3. Finally, it uses this structured knowledge to infer the answer to the original question. StructRAG outperforms existing methods on a range of challenging reasoning tasks that require combining multiple facts and insights. For example, it excels at open-ended science questions that require understanding complex processes and piecing together evidence from multiple studies. By mimicking how humans organize information to solve problems, they were able to match - or even surpass - GraphRAG while being much faster. ↓ Liked this post? Join my newsletter with 50k+ readers that breaks down all you need to know about the latest LLM research: llmwatch.com ??

BIG NEWS! ?? We're announcing the agenda for the 5th Annual MLOps World | Generative AI World Expo happening November 7th-8th in Austin, Texas. This is always an exciting/nervous time as we had to select from over 400 submissions - and it was a tough selection process, to say the least.? These are the best industry case studies, hands-on workshops, and sessions to help with your ML/AI initiatives - both high-level strategy and very deep technical dives. Selections were based off real committee pain points, so a lot of intention and curation went into each session. We'e confident you'll walk away with a ton of new resources, new skills, and will make some great connections! A huge thank you to our speakers, and everyone who submitted to speak ?? As well, to all the ridiculously skilled, creative, and innovative startups & technical teams who will be on display - we're excited to see your inspiring work. If you're working in ML/AI come join us. Come discuss your projects. You'll be in good company. See you in Austin! ?? mlopsworld .com Our Committee ?? Denys Linkov and Suhas Pai (Special thanks!**) Norm Zhou Wenjie Zi Dr. Larysa Visengeriyeva Ben Ashtiani, PhD Raphael Taiwo Alabi PhD. Nick Pogrebnyakov Shervin Shahidi Alireza Darbehani Saurabh Marathe Hien Luu Jason Carayanniotis Neeraj Madan Vincent Chio Suresh Kumar Khemka Arshad Ahmed Shailesh K. Andrea Rondelli Patrick Halina Andrew Sridhar Ashkan Amiri Vivek Murugesan Guhan Venguswamy Svetlana Zavelskaya Alex Preciado Andreea Munteanu Reza Abasi Murali Bhogavalli Nalin Dadhich Balaji Dhamodharan Tejas Chopra Alok Ranjan Pratibha Rathore Indrani Gorti Yog Seetharama Yog Seetharama Peeyush Agarwal

To facilitate Multi-Document Extraction and Graph creation, WhyHow.AI is open-sourcing Knowledge Table, an internal table-based multi-document extraction and graph creation tool, which has an agent within each cell to facilitate the extraction process. Extraction & Memory is now as easy as selecting your documents, running the set of questions you want against the system, and automatically converting the output into a set of Triples that you can immediately save as memory and query in WhyHow (or in other systems). For developers, we have found that inserting a tabular intermediary step for graph construction in your backend RAG system dramatically improves the accuracy of the graphs created. Some unique features here include: - Vector chunks tied to each cell answer - Rules & Type-based extraction guardrails - Chained Extraction Logic through Cell-to-Cell references You should use this tool if you are interested in extracting, storing and querying information across a large set of documents, as a business user or a RAG developer. Between Knowledge Table & our Platform, we provide: - Multi-Document Accuracy Uplift: 2.5x accuracy over ChatGPT 4o (in web browser) for multi-document retrieval, outperforming Text2Cypher by 2x, and beating GraphRAG. - Rule-Based Extraction Guardrails: Granular control of an open-source multi-document extraction process through Extraction Rules & Types - Ontology-Based Query Engine: An intuitive query engine that allows the user to call on both specific tools and columns directly when querying, allowing a seamless combination of both structured and unstructured retrieval Thomas Smoker Chris Rec

Thomas Smoker, cofounder/CTO of WhyHow.AI, will be speaking at the Software Lifecycle Management for Knowledge Graphs Workshop at ISWC in Baltimore in November. Catch him there! https://lnkd.in/e7k8aj5z

Thomas Smoker, cofounder/CTO of WhyHow.AI, will be speaking at the Neo4j Graph Summit tomorrow. Catch him there! https://lnkd.in/e8HZ8kED

Join us on October 1st at the Amazon Web Services (AWS) Loft in San Francisco for our October bay.area.ai meetup, the biggest, deepest, longest running and longest continuously operating AI meetup in the world. We have two deep dives in key AI deployment topics: model safety and knowledge graphs. KG-enabled AI workflows Chris Rec, WhyHow.AI In this talk, we’ll explore how knowledge graphs can supercharge your AI workflows. We'll demonstrate entity and triple extraction from semi-structured GitHub data and how to build and work with knowledge graphs using the WhyHow graph studio platform. Come discuss exciting patterns emerging in AI workflows fueled by graph technology. Chris Rec is the co-founder of WhyHow. WhyHow builds next-gen data pipelines for Knowledge Graphs to improve AI workflows. Chris is a former founder, engineer, and product manager with platform engineering experience at Netflix, AWS, and Coinbase. While AI is Moving Fast, we should Not Forget Security λ Mihai λ Maruseac λ, Google AI is moving at breakneck speed but we are also seeing security incidents, both in research papers and in real production systems. Most of these look similar to traditional software incidents of the past, but they are repeating at an accelerated pace. In this talk, we'll give an overview of why security is important for AI and how we can quickly resolve several of the most pressing issues. Mihai Maruseac is a member of Google Open Source Security team (GOSST), working on Supply Chain Security for ML. Before joining GOSST, Mihai created the TensorFlow Security team after joining Google. Previously, he worked on a startup incorporating Differential Privacy (DP) within Machine Learning (ML) algorithms (now part of Snowflake). Mihai has a PhD in Differential Privacy from UMass Boston.

What happens when your RAG system doesn’t capture the whole context, and your experts know it? Vector Only Knowledge Bases only capture what is written down. But there is so much more information trapped in the heads of your experts and employees who know the dynamic reality of what are true facts and what are the context for which these facts are used. Knowledge graph structures, augmented by LLMs, allow experts to talk and contribute directly to knowledge bases in a structured way, not just embeddings. Check out this WhyHow.AI graph structure that allows experts and users to give feedback and change the way that a RAG system stores and manipulates context as it constantly changes. Thomas Smoker Chris https://lnkd.in/eZeg8YUm