EyeLevel Launch: Your RAG is Tripping, Here's the Real Reason Why
AI Engineer World's Fair 2024 · 6:13
← All organizationswww.eyelevel.ai ↗
Enterprise document intelligence and retrieval-augmented generation
EyeLevel.ai
EyeLevel.ai builds GroundX, a platform that turns complex enterprise documents into information that AI applications can search and use. Its APIs support document ingestion, understanding and retrieval for developers building applications around private data. GroundX Studio extends this foundation with tools for developers and business users to build workflows and applications. Deployment options include managed cloud, private cloud, on-premises infrastructure and air-gapped environments, supporting organizations that need to process information within their own infrastructure.
Founded by IBM Watson veterans and developed through UC Berkeley’s SkyDeck accelerator, EyeLevel includes founder Ben Fletcher, now Valantor’s chief technology officer. GroundX integrates document parsing, chunking, retrieval representation and preparation of context for language models into one system, treating those stages as interdependent. Its engineering work includes a custom integration with DocBench for evaluating document comprehension across text, tables and graphical figures.
Valantor acquired EyeLevel in July 2026, making it a wholly owned subsidiary and incorporating GroundX into its enterprise visual-intelligence platform. Parent-company CEO Patrick McCamley leads strategy and commercialization. As of August 2026, Valantor reported more than 7,000 developers on GroundX. EyeLevel charges managed-cloud customers for ingestion, processing and search, and offers commercial on-premises deployments and consulting for organizations building document-based AI applications.
1 talk
Newest first1 speaker at AIE
Affiliations reflect their AIE appearances, not necessarily current employment.
Messages from the stage
From document extraction to reranking
Fletcher described combining fine-tuned vision models and multimodal extraction with semantic objects, multifield search, and LLM reranking instead of conventional vector-database retrieval.
Affiliations reflect each recorded session, not necessarily current employment. The talk's reported accuracy figures are company and speaker claims, not independently verified benchmarks.