# EyeLevel Launch: Your RAG is Tripping, Here's the Real Reason Why AI Engineer · 2025-02-06 Benjamin Over, co-founder of EyeLevel.ai, argues that 95% of RAG hallucinations stem from content ingestion problems, not LLMs or prompts, and introduces his platform's approach to solve this. EyeLevel avoids vector databases entirely, instead creating semantic objects with auto-generated metadata and rewriting text for search and completion. Its pipeline uses a fine-tuned vision model to extract tables, figures, and text, then applies nine fine-tuned models for ingestion and search, including a re-ranking LLM. Air France uses EyeLevel to build a call-center co-pilot, achieving over 95% accuracy on hundreds of thousands of complex documents. In a study, EyeLevel reached 98% accuracy on real-world documents, outperforming popular solutions by up to 120%. ## Questions this episode answers ### Why do RAG systems tend to produce hallucinations? According to EyeLevel co-founder Benjamin Over, hallucinations rarely stem from the LLMs or prompts; instead, they are typically caused by the RAG pipeline’s retrieval of low-quality or irrelevant content. He notes that RAG applications can have error rates up to 35% on complex enterprise documents, with three main ingestion problems: poorly extracted text, lost context from chunking, and unextracted visual elements like tables and figures. [0:54](https://aie.addtry.com/0456f874-0824-4a0e-b943-f2a8fa7c7b26?t=54000) ### What retrieval accuracy does the EyeLevel platform achieve? Benjamin Over states that EyeLevel’s customers, including Air France and Dartmouth, report their RAG applications respond correctly more than 95% of the time. A recent study found the platform reached 98% accuracy on complex real-world documents, outperforming some popular solutions by up to 120%. These results are attributed to a robust ingestion pipeline that preserves document structure and context. [2:05](https://aie.addtry.com/0456f874-0824-4a0e-b943-f2a8fa7c7b26?t=125000) ### How does EyeLevel improve RAG accuracy without using vector databases? Benjamin Over explains that EyeLevel avoids vector databases entirely, instead creating “semantic objects” that bundle original chunk text with auto-generated metadata and rewritten versions optimized for search and completion. Ingestion uses a vision model fine-tuned on millions of documents to detect images, tables, and text, plus multi-field search across the object attributes. Over nine fine-tuned models, including a reranking LLM, drive the high accuracy. [2:20](https://aie.addtry.com/0456f874-0824-4a0e-b943-f2a8fa7c7b26?t=140000) ## Key moments - **[0:00] Intro** - **[0:36] Platform** - **[0:54] Hallucinations** - [0:54] RAG applications can have error rates as high as 35%, and the source is rarely LLMs but the RAG's content ingestion problems. - **[1:45] Fix** - **[2:20] Semantic Objects** - [2:20] EyeLevel does not use vector databases and instead creates semantic objects with auto-generated metadata, achieving 98% accuracy. - [2:49] Air France uses EyeLevel to build a ChatGPT-like co-pilot for call center agents from hundreds of thousands of complex documents. - **[3:10] Air France** - **[3:32] Pipeline** - **[4:10] Structure** - **[4:32] Example** - [4:32] If you were to OCR this and extract the text from it, vectorize it, put it in your vector database, it would look something like this. - **[5:07] Search** - **[5:26] Models** - [5:26] EyeLevel's ingestion and search pipeline uses over 9 fine-tuned models to deliver production-ready RAG accuracy. - **[5:47] Invite** ## Speakers - **Benjamin Over** (guest) ## Topics RAG & Retrieval, Vision Language Models ## Mentioned Air France (company), Dartmouth (company), EyeLevel (company), The Weather Channel (company), GPT-3 (product), IBM Watson (product) ## Transcript ### Intro **Benjamin Over** [0:13] Hi, my name's Ben. I'm co-founder of EyeLevel.ai, and I've been building applications powered with AI for the last 15 years. First at IBM Research, then at IBM Watson, later working with major brands like The Weather Channel, and now at EyeLevel, where we've built the world's most accurate and scalable RAG platform. ### Platform **Benjamin Over** [0:36] Using our no-code tools and APIs, our users can upload documents and receive the most accurate retrievals in minutes. We've been developing our solution for the last 4 years, and we're among the first early users that were admitted to the GPT-3 beta program. We found it easy to get started with RAG and very difficult to master. In our own experience, RAG applications can have error or hallucination rates as high as 35%, especially when the knowledge base consists of the kinds of complicated documents that are commonly found in the enterprise. ### Hallucinations **Benjamin Over** [1:14] The source of these errors is rarely the LLMs or the prompts; instead, it's typically RAG itself, or more specifically, the quality and relevance of retrieved content. And the problems with content generally fall into one of three categories: bad or improperly extracted text, missing information from the surrounding parts of the document that's lost during chunking, or visual elements that are not extracted at all. Most commonly, the problems with RAG are content ingestion problems, and advanced RAG techniques that help you solve these problems can take hundreds of hours to implement. We've spent the last 4 years tackling these difficult data engineering problems and have built the solutions to them into our ingestion pipeline. ### Fix **Benjamin Over** [2:05] As a result, our users are able to build the most accurate RAG applications in just minutes, and our customers such as Air France and Dartmouth tell us that their RAG applications respond correctly more than 95% of the time. In a recent study, our platform achieved 98% accuracy against complicated real-world documents and outperformed some of the most popular solutions in market by as much as 120%. I'm going to quickly walk you through the unique approach we take to achieve this high level of accuracy, and I'll start by telling you that we don't use vector databases at all, and in fact, we think they may not be the best technology solution for a lot of RAG applications. ### Semantic Objects **Benjamin Over** [2:49] Instead, what we do is we create what we call semantic objects, and we do a multi-field search across the attributes of this object. I'll show you what that means with a real example from Air France. Air France has been using our platform for the last year to build a ChatGPT-like co-pilot for their call center agents. ### Air France **Benjamin Over** [3:10] They wanted to understand their knowledge base, which consists of hundreds of thousands of documents just like this one, filled with tables, figures, and text scattered across the pages. In our ingestion pipeline, the first thing we do is run a vision model that we fine-tune with millions of documents to identify where the images, the tables, and the text are. ### Pipeline **Benjamin Over** [3:32] Then we run them through dedicated multimodal processing pipelines to extract the visual and written information. When you do RAG, you have to break apart this document into smaller chunks. When you do that, you quite often lose information from around the chunks, things like what section of the document it came from, or even which document it came from. If you were to ask questions about a book and receive random paragraphs from the book, chances aren't great you'd get good answers, and that's kind of what's happening with chunking and the loss of the context. That's why we created semantic objects. ### Structure **Benjamin Over** [4:10] It consists of the original chunk text as well as auto-generated metadata that preserves the information around the text, and then we rewrite the text into two ideal formats, one for search and one for completion. Thank you. Okay, now it works better. ### Example **Benjamin Over** [4:32] Yep. Let me show you what that looks like with an example. So this is a figure from one of Air France's documents. If you were to OCR this and extract the text from it, vectorize it, put it in your vector database, it would look something like this. Look at how much information is lost in the process, though. Instead, what comes out of our ingestion pipeline is something like this. And this includes both the search version as well as the completion version of the text. When we receive a search query, we do something similar. ### Search **Benjamin Over** [5:07] We rewrite the query into a format that's compatible with the objects. Then we search the entire object, the original text, the auto-generated metadata, and the search version of the text. We use a fine-tuned LLM to re-rank the results and improve the accuracy. ### Models **Benjamin Over** [5:26] And in total, in our ingestion and search, there are more than 9 models that are fine-tuned to help deliver this kind of accuracy. The end result is the world's most accurate RAG platform, and our users are able to build enterprise-quality, production-ready applications in minutes, not months. ### Invite **Benjamin Over** [5:47] I invite you to try it for yourself, though. EyeLevel.ai/xray. Thank you very much. Thank you, Ben. Allright. --- This library is powered by PodHood (https://podhood.com), the podcast website platform.