Reduces average analysis time from 33.6 seconds to 9.8 seconds vs. existing commercial tools, boosting enterprise productivity
Elice Inc., an AI full-stack company providing AI infrastructure, cloud, and industry-specific solutions, has unveiled Helpy Document Vision, an AI document analysis solution that automatically interprets complex documents and converts them into well-structured data.
Helpy Document Vision can automatically analyze not only text paragraphs but also diverse visual elements within documents, including tables, charts, formulas, and images. Built around Helpy Table Vision, a Vision Language Model (VLM) optimized for table processing, the solution combines this proprietary model with state-of-the-art open-source models to achieve both high accuracy and high throughput.
Automatically analyzes legacy documents and large Excel files, powered by proprietary VLM “Helpy Table Vision”
Helpy Table Vision ranked No. 1 in the global VLM benchmark Nanonets and excels at analyzing complex table data used in real-world industrial settings, such as large manufacturing enterprises. With strong capabilities in domain-specific optimization, it can accurately analyze legacy documents and modern–historical records that are difficult for general-purpose models to recognize, using training data on the order of only a few hundred samples. It also supports detailed structuring tasks such as analyzing long Excel files with hundreds of rows, reconstructing chart and graph data into HTML, and converting complex formulas into symbolic representations.
In performance comparisons with existing commercial solutions, the average time required for document layout analysis and data extraction was reduced from 33.6 seconds to 9.8 seconds, making Helpy Document Vision approximately 3.4 times faster. The solution also demonstrated superior overall document understanding performance, including reading-order extraction accuracy, table and formula extraction quality, and execution time.
These results are powered by Elice Inc.’s AI full-stack capabilities built on Elice Cloud, the company’s proprietary GPU-based private infrastructure. By managing the entire workflow in-house—from infrastructure deployment to model development and service operations—Elice delivers rapid optimization and stable, high-performance services.
Helpy Document Vision can be rapidly customized for domain-specific documents in finance, healthcare, and law by tightly integrating a company’s internal data with its training infrastructure. By converting vast volumes of unstructured enterprise documents into high-quality digital data, the solution enhances the performance of RAG systems and AI agent solutions that many organizations are now adopting.
Elice Inc. also plans to extend its Vision Language Model (VLM) technology beyond document understanding into Vision-Language-Action (VLA) models capable of situational reasoning and action execution. VLA technology enables AI systems to interpret complex manuals or technical drawings and precisely control robots and machines in real industrial environments. Elice’s advanced data extraction capabilities are expected to play a key role in enabling “physical AI,” where AI agents operate directly within the physical world.
Suin Kim, Chief Research Officer (CRO) at Elice Inc., said,
“With Elice’s AI document analysis solution, we aim to help companies transform complex, manually processed documents into high-quality data so they can experience tangible innovation in workflow automation.”
Kim added,
“We will continue to advance from VLM technologies that simply read documents to VLA technologies that drive real-world actions, leading the era of ‘physical AI’ that directly solves problems on the industrial front line.”