Adit Abraham

Adit Abraham was introduced in the AI Engineer World’s Fair 2026 conference biography as co-founder and CEO of Reducto, building an AI document-intelligence platform for parsing, understanding, and structuring complex unstructured documents for AI applications. He previously studied computer science at MIT and worked on product at Google.

In his presentation, “From Ingestion to Agents: How AI Teams Build on Document Intelligence,” Abraham explains how document errors compound across an agent’s decisions. He describes combining efficient computer vision for layout detection with vision-language models for difficult content, including handwriting. His account of agentic OCR emphasizes targeted corrections—such as distinguishing zero from the letter O—while preserving source content rather than regenerating text or recalculating a document’s totals.

Abraham also presents approaches to making document structure useful for both reasoning and retrieval: Markdown for simple tables, HTML for complex tables, and separate natural-language representations for embedding models. He discusses classification and splitting to route relevant document sections, and describes his team’s use of agents with code interpreters, visualization, and repeated checks to reconstruct tables from line charts and improve structured extraction. He advocates evaluating individual pipeline stages and final agent performance, including production monitoring, and describes file-system and CLI interfaces that let agents choose document tools as needed.